Genetic Algorithm-Neural Network method

Artificial Neural Network (ANN) has several advantages; and one of the most recognized of these is the fact that it can actually learn from observing data sets. In this way, ANN is used as a random function approximation tool. These types of tools help estimate the most cost-effective and ideal methods for arriving at solutions while defining computing functions or distributions. ANN takes data samples rather than entire data sets to arrive at solutions, which saves both time and money. ANNs are considered fairly simple mathematical models to enhance existing data analysis technologies^[13].

The genetic algorithm itself can easily fall into the local minimum, slow convergence, and other shortcomings. Genetic Algorithm-Neural Network method is a simple neural network improvement. The forecast for rail freight transit is based on the Genetic Algorithm-Neural Network method. Cargo transport volume calibration is shown below (Table 6).

Table 6 Rail freight transit volume calibration

Select the transportation volume data from 2001 year to 2011 year to predict the amount of cargo transportation, 2012-2015 data is used as the validation error. To use GA to train the network in order to optimize the weights as follows:

· To determine the fitness function. The objective function that is commonly used in the network error function as a fitness function. If the error function as a fitness function, such as f (i)  1/ E (i),

(8)

i  1, 2,..., M - the number of chromosomes; k  1, 2,..., N – the learning samples; y_k – the training samples; T_k - the expected value.

· Parameter setting. Set the parameters of genetic and BP network training process, including population size, mutation probability, crossover probability, network layers, excitation functions, training, step length, accuracy and so on.

· Select a random set of weights, and calculate the value corresponding to the degree of the error and the adaptation network. The initial value of the random weights right.

· Select the largest individual genetic fitness.

· Individuals crossover and mutation operator operation, generate a new generation of groups.

· Repeat steps 3-5 until the error reaches requirement.

• Assign the step 6 result to BP neural network model for learning and prediction^[14].

Establish five input layer neurons, 15 hidden layer neurons and one output layer neurons, three-layer neural network. Specified accuracy ε  0.000001. GA initial parameters selected species populations popu  50, genetic algebra gen  100; a hidden layer activation function using tansig, the output layer activation function using purelin, training function takes train lm. And genetic algorithm combined with Matlab2012; a toolbox and artificial neural network toolbox, carried GA-BP algorithm simulation solving, and the data from 2012 year to 2015 year, as a test target. The predicted performance test results are shown in Table 7. The four-year forecast for cargo transportation at the border-crossing Dostyk – Alashankou, the data are shown below (Table 8).

Table 7 The prediction of the rail freight transit volume calibration

Table 8 Predictive value of the rail freight transit volume