Need Help Implementing Neural Network in MATLAB

Hello MATLAB community,

I am currently working on the implementation of the research paper titled "Neural Network Self-Tuning Control for a Piezoelectric Actuator" by Wenjun Li (DOI: 10.3390/s20123342).

%Equations for Weight Updates:
dW_k = eta_w * e .* Nf(y, V_k) .* Ng(y, W_k) .* (1 - Ng(y, W_k)) .* y;
dV_k = eta_v * e .* Ng(y, W_k) .* (1 - Ng(y, W_k)) .* (1 - Nf(y, V_k)) .* y;
W_k = W_k + dW_k;
V_k = V_k + dV_k;

I have provided MATLAB code snippets for weight updates in a neural network based on the equations. I'm seeking assistance in ensuring the Equations for Weight Updates are correct and also guidance on how to implement the neural network strategy described in the paper.

If anyone has experience with similar projects or is well-versed in neural networks and control systems, I would greatly appreciate your insights and advice.

Thank you in advance for your time and assistance.

[Sam Chak]

Hi @Achraf,

We could likely assist you better if you could provide us with the specific equations from the paper and explain how you derived the equations for the weight updates. Otherwise, we might need to make educated guesses, and if our guesses are incorrect, it could result in a waste of time.

[Achraf]

Hey @Sam Chak

Thank you for your response, I'm having trouble deriving the code from the equations we discussed. Your assistance in this matter would be greatly appreciated.

That's all I could achieve so far. I've had many iterations coding equation Nr17, but I didn't succeed!

num_iterations = 100; % the number of training iterations
eta_w = 0.01; % Learning rate for weights in Ng
eta_v = 0.01; % Learning rate for weights in Nf

function [W_k, V_k] = neural_network_self_tuning_control(num_iterations, eta_w, eta_v)
    % Define activation functions
    f_Ng = @(x) (1 - exp(-x)) / (1 + exp(-x)); % Sigmoid function
    f_Nf = @(x) (exp(x) - exp(-x)) / (exp(x) + exp(-x)); % Hyperbolic tangent

    % Initialize weights W_k and V_k
    W_k = rand(1, 6); % Initialize with random values
    V_k = rand(1, 6); % Initialize with random values

    % Define initial input and target values
    y_k = 5;
    yd_k = 10;

    % Training loop
    for k = 1:num_iterations
        % Calculate Ng[y(k); W(k)] and Nf[y(k); V(k)]
        Ng_yk_Wk = f_Ng(W_k * y_k);
        Nf_yk_Vk = f_Nf(V_k * y_k);

        % Calculate control input u(k)
        u_k = (yd_k - Ng_yk_Wk) / Nf_yk_Vk; %is derived from Equation (5) in the paper 

        % Update weights for Ng
        delta_w = (eta_w * f_yk / Nf_yk_Vk) * %...This must be equation 17
        W_k = W_k + delta_w;

        % Update weights for Nf
        delta_v = (eta_v * f_yk / Ng_yk_Wk) * %...This must be similar
        V_k = V_k + delta_v;

    end
end

[Sam Chak]

@Achraf, can you edit the code in your comment to list the equation numbers, like "% Refer to Eq. (Num)"? This can make the process of cross-checking the equations in the code easier and faster. Most volunteers on Sunday do not have the time to read the entire paper and find a match for each equation in the code.