Study of logistics distribution route based on improved genetic algorithm and ant colony optimization algorithm

: To solve the problem of vehicle routing problem under capacity limitation, this paper puts forward a novel method of logistics distribution route optimization based on genetic algorithm and ant colony optimization algorithm (GA-ACO). On the first stage, improved genetic algorithm with a good global optimization searching ability is used to find the feasible routes quickly. On the second stage, the result of the genetic algorithm is used as the initial solution of the ant colony algorithm to initialize the pheromone. And then improved ant colony optimization algorithm is used to find the optimal solution of logistics distribution route. Experimental results show that the optimal or nearly optimal solutions of the logistic distribution routing can be quickly obtained by this two stages method.


Introduction
For modern logistics enterprises, how to generate vehicle schedules in transportation reasonably, to optimize the transportation line, and to reduce logistics cost has become a core problem of logistics management. Due to the fact that the logistics distribution vehicle routing optimization problem is a Non-deterministic Polynomial Complete problem, using only one method to obtain the global optimal solution is difficult. So some scholars put forward particle swarm optimization combined with genetic algorithm and simulated annealing algorithm combined with genetic algorithm to solve logistics distribution route optimization problem based on the theory of combination optimization [1,2]. Genetic algorithm has the advantages of powerful global search ability and high rates of convergence, but it can't use feedback information which leads to poor search ability, premature convergence and fall into local optimum easily [3].The characteristics of ant colony algorithm is heuristic search and positive feedback mechanism, so it has the advantages of obtaining optimal solution with high efficiency, good local searching ability, distributed computing ability and strong robustness. But due to initial pheromone shortage, the solving speed is slow in early stage of searching [4]. Integrating the advantages of genetic and ant colony algorithm, this paper proposes a two stage method.

Problem Description.
Logistics distribution route optimization problem is to find the shortest total distance or lowest total freight route of transporting goods from the distribution center to multiple demanding spots by car. And it needs to satisfy the following conditions: (1) Position of each demanding spot and the demand is certain. (2) Each car has a constrain of load capacity. (3) The demand of each demanding spot must be met and be delivered only by one vehicle.

Mathematical Model.
Distribution center represented by 0 delivers goods to demanding spots with amount is N. : Constraints of each vehicle must return to distribution center and each demanding spots must be distribute by only one vehicle On the basis of meet the above constraint conditions minimizing Eq. 5, which represents the minimum path length.This is objective function.

Algorithm Step
The existing improved ant colony algorithms mainly put forward some methods for solving ant colony algorithm's problem of easily converging to local optimum, ignoring slow speed caused by initial pheromone shortage in the early searching stage. This paper proposes to use the output of improved genetic algorithm to initialize the pheromone .And making some improvement in genetic algorithm and ant colony algorithm for avoiding converging to local optimum.

1) Decoding
Based on the characteristics of logistics distribution route optimization problem, the author adopts a simple and intuitive natural number coding method [5]. With 0 representing distribution center and 1, 2,... , N representing demanding spots. The distribution paths are no more than Num which start and end at distribution center due to no more than Num vehicles in distribution center. In order to reflect the paths of the vehicle distribution in the code, increasing the number of Num-1 virtual distribution centers respectively represented by Num+1,Num+2,...,Num+N-1. Thus, a random arrangement of 1 to Num+N-1 represents an individual corresponding to a distribution plan.
For example, a distribution center distributes goods to 8 demanding spots with no more than 3 vehicles. 1 to 8 represent demanding spots and 9 to 10 represent virtual distribution centers. A random arrangement of 1 to 10 represents logistics distribution plan.
2) Generate initial population Randomly generating an arrangement of 1 to Num+N-1 forms an individual. In this way, a initial population is generated.
3) Evaluate fitness The constrains may not be satisfying after steps of genetic operations. So, the fitness function must reflect the feasibility and the cost of corresponding solution.
For individual r, the number of unfeasible path in corresponding distribution route scheme is represented by M(If M equals 0, the individual is a feasible solution).Then the fitness function F is: C is the value of Eq. 5 . G is a punishment weight for each of the unfeasible path, taking a relatively large positive number according to the scope of the objective function. 4)Select Select optimal and sub-optimal individual measuring by fitness function. 5)Crossover I adopt a method similar to OX [6] called sequence reversal crossover operator. Compared with other methods, this method can produce a certain degree of variation in the situation of same parents, which has a certain effect on the diversity of the population.
Step D: Reverse the remaining S3' arrangement in addition to the part of Q2 , the similar operation on S4'.

13
The 2 randomly generated numbers are 4 and 8:

6)Mutation
The purpose is to dig out the diversity of individuals in the population, and to overcome the disadvantages of genetic manipulation which may be limited to local solutions.
Mutation is occurred with probability of P. If mutation happens, the exchange number is randomly generated.
If the mutation operation occurs, and the switching frequency J = 3; 8) Select the best of the previous r individuals measured by Eq. 5 in population as the initial input of the latter algorithm through decoding.

Improved Ant Colony Algorithm
(1) Initialization Set initial pheromone. Set the initial r paths of ant colony algorithm by the output of genetic algorithm.Using the Eq. 7 to initial pheromone. Set the maximum iterative algebra and the number of ants k;  is the concentration of pheromone on the path of i to j (2)Tectonic solution Firstly, according to the Eq. 10 to determine the next point. The traditional ant colony algorithm based on the probability calculated by Eq. 11 in accordance with the roulette wheel method generates the next access point.The improved algorithm introduces a deterministic search and uncertain search. The deterministic search uses the gained experience to guide the path selection, making up for the defects of exploratory search restricted in convergence speed. Through appropriate adjustment of 1 q , making the deterministic search and exploratory search reasonable collocation can accelerate the convergence speed of ACA.
q is a constant between 0 to 1. 2 q is a random number between 0 to 1. Allowed(k) is the collection of demanding spots that are not visited.α is the heuristic information coefficient, which indicates the relative importance of trace amount of residual information. The greater the value, the more inclined the ants tend to choose the path of other ants ;β is a parameter to control the influence of visibility.The greater the value, the more close it is to the rule of greedy; j i, d is the distance from i to j.
This paper also optimizes the heuristic function Secondly, determine whether to meet the requirements if access this point: If satisfied, it will be directly added to the current path; If it doesn't satisfied, the ant go back to the distribution center, then start from the distribution center to reach the location. Finally remove the demanding spot from Allowed (k).
Thirdly, judge whether the Allowed (k) is empty:If it is empty, execute 2.4; If not, execute 2.2. Finally, update local pheromone using Eq. 14 (3) By using 2-OPT sub_routes optimization, the solution of each iteration process is improved, the length of the optimal solution is shortened and the rate of convergence algorithm is improved.
(4) Upd Choose optimal so pheromone solution of

Analysis
According to the characteristics of logistics distribution path optimization problem, this paper proposes an optimal path algorithm based on GA-ACO. By introducing crossover and mutation operator, the algorithm can avoid the premature and stagnation of the algorithm in the local search process. The improvements in initial pheromone assigned by genetic algorithm results, the strategy of updating pheromone and on the choice of selecting next demanding spot of ACO enhance the positive feedback effect, thus improving the convergence speed and the global searching ability of the algorithm. The experimental results show that the improved GA-ACO can quickly and effectively obtain the optimal solution or approximate optimal solution. It has a certain reference value for the study of the optimization of logistics distribution routing problem.