11.2 Theory of ANNPARTOutputActivation FunctionhardlimpurelinlogsigHiddenInputHiddenOuterLayerLayerLayerLayer10=1+e-net1Vx1net =wixi+.x2.i=0x3x4WoWiW2WnXo=1XXInputx6西步交通大学
6 PART 1 w1 w2 wn w0 x0=1 Output x1 x2 xn . . . Input x j n i net = wixi + =0 net e o − + = 1 1 Activation Function 1.2 Theory of ANN
T1.3 Types of ANNPARTPercentagesofdifferenttypesofANNsBPNusedinnuclearresearchesSOM57.14%PNNGRNNBPNGNNSOMPNNGMDH-ANN7.14%GRNNRBFNGNN1.02%GMDH-ANNANFISRBFN4.08%WNNANFIS2.04%WNNSONN8.16%SONN3.06%1.02%HONN5.1%HONN8.16%1.02%1.02%FCNNFCNN1.02%typenotgiven西步交通大学
7 PART 1 1.3 Types of ANN ➢ BPN ➢ SOM ➢ PNN ➢ GRNN ➢ GNN ➢ GMDH-ANN ➢ RBFN ➢ ANFIS ➢ WNN ➢ SONN ➢ HONN ➢ FCNN ➢ . Percentages of different types of ANNs used in nuclear researches
11.3 Types of ANNPARTBPN (Back PropagationNeural Network)is the mostwidelyusedtypeinnuclearengineeringareaShortcomingsofBPNDifficult to determine the training parameters, such as the面number of hidden layers and number of neuron;II. Relatively low computation efficiency and rate of convergencetime-consuming training;IIl. Probable stop of the computation when reaching the localminimum errorGenetic Algorithm (GA)GNNisan effectivemethod tooptimize BPN8西步交通大学
8 PART 1 1.3 Types of ANN ⚫ Shortcomings of BPN: I. Difficult to determine the training parameters, such as the number of hidden layers and number of neuron; II. Relatively low computation efficiency and rate of convergence, time-consuming training; III. Probable stop of the computation when reaching the local minimum error Genetic Algorithm (GA) is an effective method to optimize BPN BPN (Back Propagation Neural Network) is the most widely used type in nuclear engineering area GNN
11.3 Types of ANNPARTGNN(GeneticNeural Network)usesGeneticAlgorithmtooptimizeBPNStartGeneticAlgorithmCrossoverProblemAnalysisSGA=(C,Ef,Pmi,M,Φ,T,Y,TMutationNeural networkmodelevaluationselectionNConvergencePopulationTXinitializationTrainingFitnesscalculationEndSelectionmutationcrossover9西步交通大学
9 PART 1 1.3 Types of ANN GNN (Genetic Neural Network) uses Genetic Algorithm to optimize BPN SGA C E P M T = ( , , , , , , , f ini e ) Genetic Algorithm Start Problem Analysis Neural network model Population initialization Fitness calculation Selection Crossover Mutation Convergence Training End Y N
西安交通大学XIANJIAOTONG UNIVERSITY02ANNApplications inT/HProblems1896
ANN Applications in T/H Problems