A DDoS-Attack Detection Method Oriented to the Blockchain Network Layer
Algorithm 1
Parametric training algorithm for the CMCNN model.
Input: Training set Dtrain, CMCNN model f, the number of neural network layers L, bias b, learning rate h, momentum parameter a, and batch training size batch_size. Here, Es(θ) is the loss function of the regularisation term with and b; δ(N) is the overall loss value in each training round of the CMCNN model. Z(l)(l) is the input in the lth layer; and bf are the weight matrix and bias of the previous round of iteration, respectively; ε is the threshold value of changes in accuracy rate; and ΔAccuracy is the accuracy of the trained model.
Output: Parameter of the CMCNN model
Start
(1)
while ΔAccuracy > e && δ(N) < 0.2
(2)
Dsample = Sampling(Dtrain, M)//Generate training sample with a batch size of M
(3)
yt = Label (Dsample)//Generate dataset label
(4)
x = Data (Dsample)//Generate dataset to be trained
(5)
θCNN←ParaInitCNN //Initialise parameters of the CNN module
(6)
θSSAE←ParaInitXavier //Initialise the parameters of the SAE module
(7)
for l ← L − 1 to 0 do://Train the CMCNN model layer-by-layer
(8)
← (wl, bl)CMCNN//Update parameters in CNN and SAE modules layer-by-layer
(9)
//Compute overall error of the CMCNN model
(10)
δ(l + 1) ← //Compute the error of each layer of the model
(11)
//Adaptively adjust ISGD learning rate
(12)
//Update weight parameter according to the ISGD method
(13)
//Update bias parameter according to the ISGD method
(14)
end for
(15)
ΔAccuracy = d(N) − d(N − 1)//Compute changes in loss value
