电池SOH仿真系列-基于LSTM神经网络的电池SOH估算方法

基于 LSTM神经网络的电池SOH估算

  循环神经网络（Recurrent Neural Network，RNN）与BP神经网络不同，RNN网络不仅考虑前一时刻的输入，同时还赋予网络对前面时刻信息的记忆能力。尽管RNN网络具有较高的精度，但其存在着梯度消失的问题。对此，出现了一系列改进的RNN网络，而LSTM神经网络就是其中改进效果最好的一种。基于LSTM神经网络的电池SOH估算方法具体如下所示：
  （1）锂离子电池循环寿命数据
  数据来源于NASA研究中心搭建的锂离子电池测试平台，选取5号锂离子电池（额定容量为2Ah）。循环测试实验在室温下进行：以1.5A的恒流电流为电池充电，直到充电截止电压（4.2V），以恒压电流充电，直到充电电流降至20mA；在2A恒流（CC）模式下放电，直至电池分别降至2.7V、2.5V、2.2V、2.5V。当电池达到EOL（end-of-life, EOL）标准时，实验停止，额定容量下降30%。
  （2）数据预处理
  利用最大最小值归一化方法将数据放缩到0和1之间。

  式中，xmax是输入数据x的最大值，xmin是输入数据x的最小值。
  （3）仿真分析
  将NASA锂离子电池实验数据中的充电容量作为模型输入，模型的输出为以放电容量为参考的电池 SOH。经过训练，神经网络可以迭代得到网络权重及偏置具体参数。

  从图中可以看出，模型对于电池容量衰减的整体趋势预测是非常精准的。其具体代码如下所示：

function pre_data = LSTM_main(d,h,train_data,test_data)
%% 预处理
lag = 8;
% d = 51;
[train_input,train_output] = LSTM_data_process(d,train_data,lag); %数据处理

[train_input,min_input,max_input,train_output,min_output,max_output] = premnmx(train_input',train_output');

input_length = size(train_input,1); %样本输入长度
output_length = size(train_output,1); %样本输出长度
train_num = size(train_input,2); %训练样本个数
test_num = size(test_data,2); %测试样本个数
%% 网络参数初始化
% 结点数设置
input_num = input_length;
cell_num = 10;
output_num = output_length;
% 网络中门的偏置
bias_input_gate = rand(1,cell_num);
bias_forget_gate = rand(1,cell_num);
bias_output_gate = rand(1,cell_num);
%网络权重初始化
ab = 20;
weight_input_x = rand(input_num,cell_num)/ab;  
weight_input_h = rand(output_num,cell_num)/ab;  
weight_inputgate_x = rand(input_num,cell_num)/ab;  
weight_inputgate_h = rand(cell_num,cell_num)/ab;  
weight_forgetgate_x = rand(input_num,cell_num)/ab;  
weight_forgetgate_h = rand(cell_num,cell_num)/ab;  
weight_outputgate_x = rand(input_num,cell_num)/ab;  
weight_outputgate_h = rand(cell_num,cell_num)/ab;  
%hidden_output权重
weight_preh_h = rand(cell_num,output_num);
%网络状态初始化
cost_gate = 1e-6;
h_state = rand(output_num,train_num+test_num);
cell_state = rand(cell_num,train_num+test_num);
%% 网络训练学习
for iter = 1:3000 %迭代次数
    yita = 0.01;  %每次迭代权重调整比例
    for m = 1:train_num
    
        %前馈部分
        if(m==1)
            gate = tanh(train_input(:,m)' * weight_input_x);
            input_gate_input = train_input(:,m)' * weight_inputgate_x + bias_input_gate;
            output_gate_input = train_input(:,m)' * weight_outputgate_x + bias_output_gate;
            for n = 1:cell_num
                input_gate(1,n) = 1 / (1 + exp(-input_gate_input(1,n)));%输入门
                output_gate(1,n) = 1 / (1 + exp(-output_gate_input(1,n)));%输出门
                %sigmoid函数
            end
            forget_gate = zeros(1,cell_num);
            forget_gate_input = zeros(1,cell_num);
            cell_state(:,m) = (input_gate .* gate)';
            
        else
            gate = tanh(train_input(:,m)' * weight_input_x + h_state(:,m-1)' * weight_input_h);
            input_gate_input = train_input(:,m)' * weight_inputgate_x + cell_state(:,m-1)' * weight_inputgate_h + bias_input_gate;
            forget_gate_input = train_input(:,m)' * weight_forgetgate_x + cell_state(:,m-1)' * weight_forgetgate_h + bias_forget_gate;
            output_gate_input = train_input(:,m)' * weight_outputgate_x + cell_state(:,m-1)' * weight_outputgate_h + bias_output_gate;
            for n = 1:cell_num
                input_gate(1,n) = 1/(1+exp(-input_gate_input(1,n)));
                forget_gate(1,n) = 1/(1+exp(-forget_gate_input(1,n)));
                output_gate(1,n) = 1/(1+exp(-output_gate_input(1,n)));
            end
            cell_state(:,m) = (input_gate .* gate + cell_state(:,m-1)' .* forget_gate)';   
        end
  
        pre_h_state = tanh(cell_state(:,m)') .* output_gate;
        h_state(:,m) = (pre_h_state * weight_preh_h)';
        %误差计算
        Error = h_state(:,m) - train_output(:,m);
        Error_Cost(1,iter)=sum(Error.^2);  %误差的平方和（4个点的平方和）
        if(Error_Cost(1,iter)<cost_gate)   %判断是否满足误差最小条件
            flag = 1;
            break;
        else     
            [  weight_input_x,...
               weight_input_h,...
               weight_inputgate_x,...
               weight_inputgate_h,...
               weight_forgetgate_x,...
               weight_forgetgate_h,...
               weight_outputgate_x,...
               weight_outputgate_h,...
               weight_preh_h ] = LSTM_updata_weight(m,yita,Error,...
                                                   weight_input_x,...
                                                   weight_input_h,...
                                                   weight_inputgate_x,...
                                                   weight_inputgate_h,...
                                                   weight_forgetgate_x,...
                                                   weight_forgetgate_h,...
                                                   weight_outputgate_x,...
                                                   weight_outputgate_h,...
                                                   weight_preh_h,...
                                                   cell_state,h_state,...
                                                   input_gate,forget_gate,...
                                                   output_gate,gate,...
                                                   train_input,pre_h_state,...
                                                   input_gate_input,...
                                                   output_gate_input,...
                                                   forget_gate_input,input_num,cell_num);
       
        end
    end
    if(Error_Cost(1,iter)<cost_gate)
        break;
    end
end

%% 测试阶段
%数据加载
test_input = train_data(end-lag+1:end);
test_input = tramnmx(test_input',min_input,max_input);
% test_input = mapminmax('apply',test_input',ps_input);

%前馈
for m = train_num + 1:train_num + test_num
    gate = tanh(test_input' * weight_input_x + h_state(:,m-1)' * weight_input_h);
    input_gate_input = test_input' * weight_inputgate_x + h_state(:,m-1)' * weight_inputgate_h + bias_input_gate;
    forget_gate_input = test_input' * weight_forgetgate_x + h_state(:,m-1)' * weight_forgetgate_h + bias_forget_gate;
    output_gate_input = test_input' * weight_outputgate_x + h_state(:,m-1)' * weight_outputgate_h + bias_output_gate;
    for n = 1:cell_num
        input_gate(1,n) = 1/(1+exp(-input_gate_input(1,n)));
        forget_gate(1,n) = 1/(1+exp(-forget_gate_input(1,n)));
        output_gate(1,n) = 1/(1+exp(-output_gate_input(1,n)));  
    end
    cell_state(:,m) = (input_gate .* gate + cell_state(:,m-1)' .* forget_gate)';  
    pre_h_state = tanh(cell_state(:,m)') .* output_gate;
    h_state(:,m) = (pre_h_state * weight_preh_h)';
    
    % 将当前预测点作为下一步输入数据
    test_input = postmnmx(test_input,min_input,max_input);
    now_prepoint = postmnmx(h_state(:,m),min_output,max_output);
    %test_input = mapminmax('reverse',test_input,ps_input);
    test_input = [test_input(2:end); now_prepoint];
    test_input = tramnmx(test_input,min_input,max_input);
end

pre_data = postmnmx(h_state(:,train_num + h:h:train_num + test_num),min_output,max_output);

all_pre = postmnmx(h_state(:,1:h:train_num + test_num),min_output,max_output);

% 画图
figure
title('LSTM预测')
hold on
plot(1:size([train_data test_data],2),[train_data test_data], 'o-', 'color','r', 'linewidth', 1);
plot(size(train_data,2) + h:h:size([train_data test_data],2),pre_data, '*-','color','b','linewidth', 1);
plot([size(train_data,2) size(train_data,2)],[-0.01 0.01],'g-','LineWidth',4);
legend({ '真实值', '预测值'});
end

  想了解更过相关仿真，可以关注我的微信公众号。