Matlab基于QRCNN-Attention多变量时序预测 基于分位数回归的卷积神经网络结合注意力机制的多变量时序预测

效果分析

基本介绍

QRCNN-Attention多变量时序预测 基于分位数回归的卷积神经网络结合注意力机制的多变量时序预测。运行环境matlab2021及以上

直接替换Excel数据即可用!注释清晰，适合新手小白~

订阅专栏只能获取专栏内一份代码。

程序设计

function pimwp = PINAW(T_sim, T_train)

%%  矩阵转置
if size(T_sim, 1) ~= size(T_train, 1)
    T_sim = T_sim';
end

%%  区间平均宽度百分比
pimwp = 1 / length(T_train) * sum((T_sim(:, end) - T_sim(:, 1))...
                                ./ (max(T_train)-min(T_train)));

end

function [in,out]=data_process(data,num)
% 采用1-num作为输入 第num+1作为输出
n=length(data)-num;
for i=1:n
    x(i,:)=data(i:i+num);
end
in=x(:,1:end-1);
out=x(:,end);

function [mae,rmse,mape,error]=calc_error(x1,x2)

error=x2-x1;  %计算误差
rmse=sqrt(mean(error.^2));
disp(['1.均方差(MSE)：',num2str(mse(x1-x2))])
disp(['2.根均方差(RMSE)：',num2str(rmse)])

 mae=mean(abs(error));
disp(['3.平均绝对误差（MAE）：',num2str(mae)])

 mape=mean(abs(error)./x1);
 disp(['4.平均相对百分误差（MAPE）：',num2str(mape*100),'%'])
Rsq1 = 1 - sum((x1 - x2).^2)/sum((x1 - mean(x2)).^2);
disp(['5.R2：',num2str(Rsq1*100),'%'])
end




classdef spatialDropoutLayer < nnet.layer.Layer & nnet.layer.Formattable
    % Example custom spatial dropout layer.

    properties
        DropoutFactor
    end

    methods
        function layer = spatialDropoutLayer(dropoutFactor,NameValueArgs)
            % layer = spatialDropoutLayer creates a spatial dropout layer
            % with dropout factor 0.02;
            %
            % layer = spatialDropoutLayer(dropoutProb) creates a spatial
            % dropout layer with the specified probability.
            %
            % layer = spatialDropoutLayer(__,Name=name) also specifies the
            % layer name using any of the previous syntaxes.

            % Parse input arguments.
            arguments
                dropoutFactor = 0.02;
                NameValueArgs.Name = ""
            end
            name = NameValueArgs.Name;

            % Set layer properties.
            layer.Name = name;
            layer.Description = "Spatial dropout with factor " + dropoutFactor;
            layer.Type = "Spatial Dropout";
            layer.DropoutFactor = dropoutFactor;
        end

        function Z = predict(layer, X)
            % Forward input data through the layer at prediction time and
            % output the result.
            %
            % Inputs:
            %         layer - Layer to forward propagate through 
            %         X     - Input data
            % Output:
            %         Z     - Output of layer forward function

            % At prediction time, the output is unchanged.
            Z = X;
        end

        function Z = forward(layer, X)
            % Forward input data through the layer at training
            % time and output the result and a memory value.
            %
            % Inputs:
            %         layer - Layer to forward propagate through 
            %         X     - Input data
            % Output:
            %         Z - Output of layer forward function

            dropoutFactor = layer.DropoutFactor;

            % Mask dimensions.
            fmt = dims(X);
            maskSize = size(X);
            maskSize(ismember(fmt,'ST')) = 1;

            % Create mask.
            dropoutScaleFactor = single(1 - dropoutFactor);
            dropoutMask = (rand(maskSize,'like',X) > dropoutFactor) / dropoutScaleFactor;

            % Dropout.
            Z = X .* dropoutMask;
        end
    end
end

参考资料

[1] https://blog.csdn.net/m0_57362105/article/details/128364733?spm=1001.2014.3001.5501
[2] https://blog.csdn.net/m0_57362105/article/details/128327690?spm=1001.2014.3001.5502