Cold-chain logistics distribution path optimization system and method based on deep neural network and improved discrete Hemma algorithm

Abstract

The invention relates to a cold-chain logistics distribution path optimization system and method based on a deep neural network and an improved discrete Hemma algorithm, and belongs to the field of operation planning optimization and intelligent scheduling. Firstly, an initial population is generated by adopting chaotic mapping; traversing all individuals of the population in each iteration, moving a current solution to an optimal solution by using improved discrete river horse movement in the initial stage of search, quickly converging the current solution to the optimal solution by using a partial matching operator in the later stage of the algorithm, and generating a random new solution with a certain probability; and then traversing the other half of individuals to generate a random new solution, comparing the fitness of the new solution and the fitness of the current solution, updating the current solution according to a comparison result, then traversing the whole group solution again, performing a VNS operator on each solution to update the solution, and finally adding a jump-out local optimal insurance mechanism. According to the invention, the logistics distribution efficiency can be effectively improved, and the time and cost in the cold-chain logistics process can be reduced.

Description

Technical Field

[0001] The present invention relates to a cold chain logistics distribution path optimization system and method based on a deep neural network and an improved discrete Hippo algorithm, and belongs to the field of operations optimization and intelligent scheduling. Background Art

[0002] In recent years, with the rapid advancement of internet technology and the continuous improvement of people's quality of life, people's lifestyles have become more modern and convenient, and their quality of life has continued to improve. This has led to a particularly high demand for fresh produce, particularly fresh flowers and food. The growing demand for low-temperature, fresh-keeping logistics and delivery services for this type of express delivery has driven the booming fresh produce logistics sector. However, the impact of cold chain logistics on global warming urgently needs to be addressed. Carbon emissions from the transportation industry account for over 10% of global emissions, primarily from transportation. Rising living standards have driven a rapid increase in the number of motor vehicles. Even with a well-connected urban road network and relatively comprehensive road infrastructure, this growth cannot be met. Therefore, simply building urban subways and light rail transit will not completely solve urban road congestion. By combining new technologies such as communications, computers, global positioning systems, and geographic information systems, urban traffic management can be provided dynamically and efficiently with rich, real-time traffic information. By rapidly processing and utilizing this massive amount of data, drivers can avoid peak travel times, develop appropriate delivery routes, and develop contingency plans. This can not only reduce vehicle dwell time but also ensure a balanced distribution of traffic across the road network. Therefore, optimizing logistics distribution routes based on real-time traffic information can positively contribute to alleviating urban traffic congestion. In cold chain logistics and distribution scenarios, cold chain distribution serves as the "last mile" of the logistics industry. Reasonable and efficient route planning can effectively reduce travel time for refrigerated vehicles and lower various costs in the process. However, vehicle routing problems are NP-hard problems characterized by large scale, numerous constraints, and high solution difficulty. Existing research on this type of problem primarily employs heuristic rule-based methods and mathematical optimization methods. Heuristic methods, such as "hill climbing," are common in practical applications due to their simplicity and ability to produce high-quality solutions within a limited timeframe. However, they are prone to falling into local optima, resulting in low solution efficiency and unsuitability for large-scale problems. Mathematical optimization methods, such as integer programming and mixed integer linear programming, can achieve relatively optimal solutions for problems of a certain scale. However, as the scheduling scale increases, the computational complexity increases significantly, making them difficult to meet the scheduling efficiency and real-time requirements in real-world scenarios. In cold chain logistics and distribution, multiple factors, such as service time windows, the number of vehicles used, and delivery route planning, further increase the complexity of the problem. However, traditional methods have significant limitations in dealing with such complex problems. Therefore, it is necessary to explore more efficient and practical solutions. Summary of the Invention

[0003] The technical problem solved by the present invention is to provide a cold chain logistics distribution path optimization system and method based on deep neural network and improved discrete Hippo algorithm, which effectively reduces the time cost, fuel consumption, flower decay cost, refrigeration cost, carbon emission cost, and time penalty cost in the distribution process.

[0004] The technical solution adopted by the present invention is: a cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm, including:

[0005] Flowers, goods that require delivery vehicles to complete delivery;

[0006] The client, the recipient of the flowers;

[0007] Refrigerated warehouses for storing flowers that need to be delivered by refrigerated vehicles;

[0008] Refrigerated vehicles are used to deliver flowers to customers at designated locations and complete unloading;

[0009] The intelligent dispatching system is used to dispatch refrigerated vehicles to deliver flowers from the refrigerated warehouse to the customer's designated location based on the customer's appointment.

[0010] Preferably, the fresh flowers are easily perishable, easily withered and difficult to preserve for a long time.

[0011] Preferably, at the time of dispatch initiation, all refrigerated vehicles are available and departing from the refrigerated warehouse location.

[0012] Preferably, the service time window reserved by the customer with the intelligent scheduling system should be greater than or equal to the time required for the refrigerated vehicle to unload the goods.

[0013] Preferably, once the unloading operation of the refrigerated vehicle at the customer's place begins, no interruption is allowed.

[0014] Preferably, each customer can only be served once by a refrigerated vehicle.

[0015] A cold chain logistics distribution path optimization method based on deep neural network and improved discrete Hippo algorithm, the specific steps are as follows:

[0016] Step 1: Use natural number encoding to represent each individual, where each individual corresponds to a solution. The structure of the solution represents the delivery order sequence of customers. Assuming that the number of customers who need to unload is n, the length of the solution is n;

[0017] Step 2: Use the chaotic mapping method to generate the initial solution. This method first obtains a new solution for the population size through circle chaotic mapping, then maps it to an integer sequence with a maximum value equal to the number of customers n, and finally checks and replaces repeated numbers.

[0018] Step 3: Use a deep neural network to predict the traffic speed in the delivery area. Based on the traffic speed, the time cost, fuel consumption, flower decay cost, cooling cost, carbon emission cost, and time penalty cost are calculated by taking into account the customer's task volume and location. An optimization objective function is constructed to minimize the overall cost. The inverse of this objective function is used as the algorithm's fitness function. A larger fitness value indicates a better solution performance.

[0019] Step 4: Traverse half of the individuals in the population. During the traversal process, three different processing solutions are selected through an adaptive parameter T that changes with the number of iterations. The first one is that when the value of T is less than 0.6, the current traversal solution uses an improved discrete hippopotamus movement method to move the current solution toward the optimal solution. The principle is to compare the current solution with the optimal solution bit by bit. If the current position is different, the current solution is modified. At the same time, the current solution is traversed bit by bit to find the same code as the modified value, and it is modified to the value before the modification at the modified position. The current solution is traversed bit by bit in this way: the second one is that when the value of T is greater than 0.6, and if a randomly generated If the value of T is greater than 0.5, a partial matching crossover operation is used to move and transform the individuals in the current population; the partial matching crossover operation refers to selecting part of the code from the parent generation and directly retaining it in the offspring to generate a new solution; the third method is that when the value of T is greater than 0.6, and if a randomly generated value is less than 0.5, a non-repeating sequence from 1 to n is randomly generated to simulate the large-scale random movement of individuals in the population: then one of the methods is selected to move the individuals, and the fitness before and after the transformation is compared to decide whether to apply the new solution. According to the above rules, half of the population is traversed to complete the first stage of the improved Hippo algorithm: the exploration stage;

[0020] Step 5: Then enter the second stage of the improved hippopotamus optimization algorithm: defense stage: traverse the other half of the individuals in the population, randomly generate a new solution and compare the fitness with the current solution, simulate the hippopotamus defense predator stage in nature, when the predator's fitness is greater than the current solution, move toward the predator by improving the discrete hippopotamus movement method, simulate the natural hippopotamus taking a step forward to confront the hippopotamus, when the predator's fitness is greater than the current solution, use multiple 2-opt operations to move toward the optimal solution leading hippo in the population, and decide whether to apply the new solution by comparing the fitness before and after the transformation, and traverse the other half of the population according to the above rules.

[0021] Step 6: Then enter the third phase of the improved Hippo optimization algorithm: the defense phase: re-traverse all individuals in the population. Each traversal performs a VNS search on the current individual. The VNS search operator uses multiple random deletion greedy insertion operators;

[0022] Step 7: Based on the solutions obtained in Steps 4, 5, and 6, an insurance mechanism is added to prevent the population from falling into the local optimum. The mechanism uses a random reversal mutation operator and applies the Metripolis criterion to make the hippopotamus randomly move out of the local optimum. It detects whether the current iterative individual solution has not been updated for more than five times. If the solution has not been updated for many times, the mutation operator is used to mutate the current solution multiple times, and the Metripolis criterion is used to accept the poor solution in the hope of obtaining a better solution.

[0023] The beneficial effects of the present invention are: taking into full consideration key information such as the service time window of the customer's appointment, the customer's location, and the customer's cargo demand, a platform scheduling optimization model is constructed with the goal of minimizing time cost, fuel consumption, flower decay cost, refrigeration cost, fuel consumption, carbon emission cost, and time penalty cost. Since the optimization goal involves multiple complex factors, the problem is difficult to solve. Therefore, a method based on a deep neural network and an improved discrete Hippo algorithm is proposed. The algorithm has significant advantages in terms of solution quality, convergence speed, etc., and can effectively improve the efficiency and stability of solving complex optimization problems. Through the combination of this model and the algorithm, the cost loss in the cold chain distribution process can be significantly reduced, and the time waste in the cold chain distribution logistics process can be reduced, thereby achieving the optimization of path planning, reducing operating costs and improving economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the system model of the present invention;

[0025] Figure 2 It is a schematic diagram of the individual coding solution of the present invention;

[0026] Figure 3 is a schematic diagram of the deep neural network prediction model of the present invention;

[0027] Figure 4 It is a schematic diagram of the improved discrete hippopotamus movement operation of the present invention;

[0028] Figure 5 is a schematic diagram of a partial matching crossover operation of the present invention;

[0029] Figure 6 It is a schematic diagram of multiple 2-opt operations of the present invention;

[0030] Figure 7 This is a schematic diagram of the operation of random deletion and greedy insertion of the present invention;

[0031] Figure 8 is a schematic diagram of the inversion mutation operation of the present invention;

[0032] Figure 9 It is a schematic flow chart of the solution algorithm of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0034] Example 1: The present invention fully considers the road traffic conditions in the distribution area, the customer's service time window, the customer's location and other information, and aims to reasonably allocate distribution routes for vehicles, and constructs a cold chain logistics distribution intelligent scheduling road optimization model with the goal of minimizing delivery time, fuel consumption cost, flower loss, time penalty cost, and unreasonable routes that lead to additional refrigeration, carbon emissions and other costs. In order to solve the model, a solution based on deep neural network and improved discrete hippo algorithm is proposed. The algorithm first uses chaotic mapping to generate uniformly dispersed solutions in the high-dimensional solution space to enhance the breadth of exploration. Then, half of the hippos in the population are explored and optimized in turn, and different exploration actions are used using the adaptive variable T. As the number of iterations increases, the algorithm dynamically adapts to the early search width and the later search speed requirements, uses a partial matching operator to update the hippo position, and uses a Cauchy mutation with a long-tail effect to enhance the randomness of the exploration. Then, the optimization is performed on the other half of the population, and a new random solution is generated in the solution space, named predator. The fitness of the current hippopotamus individual and the predator are compared. If the predator is better, the discretized hippopotamus position update is enabled, and the hippopotamus moves towards the better predator. Otherwise, the hippopotamus moves randomly. Finally, the escape phase of the hippopotamus is entered: the entire hippopotamus population is migrated, local optimization is performed, and convergence is accelerated. The VNS search operator uses an improved destruction and repair operator. Finally, an insurance mechanism to prevent local optimality is added: in order to prevent the solution from being trapped in the local optimality for a long time, an insurance mechanism is designed. When a hippopotamus individual does not move for a long time, the present invention uses a mutation operator and applies the Metripolis criterion to make the hippopotamus randomly move out of the local optimality to avoid premature convergence, thereby improving the convergence efficiency and the accuracy of the solution, and reducing the time waste and resource consumption in the cold chain logistics distribution process.

[0035] like Figure 1 As shown, a cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm includes:

[0036] Flowers, goods that require delivery vehicles to complete delivery;

[0037] The client, the recipient of the flowers;

[0038] Refrigerated warehouses for storing flowers that need to be delivered by refrigerated vehicles;

[0039] Refrigerated vehicles are used to deliver flowers to customers at designated locations and complete unloading;

[0040] The intelligent dispatching system is used to dispatch refrigerated vehicles to deliver flowers from the refrigerated warehouse to the customer's designated location based on the customer's appointment.

[0041] Furthermore, the flowers are prone to rotting and withering and are difficult to preserve for a long time.

[0042] Furthermore, at the initial time of scheduling, all refrigerated vehicles are available and depart from the refrigerated warehouse location.

[0043] Furthermore, the service time window reserved by the customer with the intelligent scheduling system must be greater than or equal to the time required for the refrigerated vehicle to unload the goods.

[0044] Furthermore, once the unloading operation of the refrigerated vehicle at the customer's site begins, it is not allowed to be interrupted.

[0045] Further, each customer can only be served once by a refrigerated vehicle.

[0046] like Figure 2 As shown, natural number encoding is used, where each individual corresponds to a solution. The structure of the solution represents the order in which customers are delivered. Assume there are eight customers with orders, numbered 1-8. For example, the solution is 8, 1, 5, 2, 4, 3, 6, 7. This means the order in which customers are delivered is 8, 1, 5, 2, 4, 3, 6, 7. The length of the solution is equal to the number of customers requiring delivery. If the number of customers requiring delivery is n, then the length of the solution is n.

[0047] like Figure 3 As shown in the figure, a traffic speed prediction model is constructed using a deep neural network. First, based on a dataset of historical real-world vehicle speeds, an enhanced matrix module is constructed by incorporating attributes such as point of interest (POI) and weather and holiday information from road nodes. The model then predicts vehicle speeds at road nodes based on the spatial dependency of roads and the temporal dependency of speed data. Graph neural networks and recurrent neural networks are used to extract spatiotemporal features, combined with an attention mechanism, to predict traffic speeds through a fully connected layer. This prediction is then embedded in an improved discrete Hippo optimization algorithm.

[0048] like Figure 3As shown in the figure, the core idea of improving the discrete hippopotamus movement operation is to compare each bit of the current individual with a target individual. If it is different from the current position, the current solution is modified with a certain probability. At the same time, traverse each bit of the current solution to find the encoding that is the same as the modified value, and modify it to the number before the modification at the modified position. Traverse the current solution bit by bit in this way to generate a new solution. The specific operation steps are as follows: First, select two individuals, denoted as the current solution and the target solution. Assume the length of the solution is 8. Compare the current solution with the target solution bit by bit. If the customer encodings at the current position of the two solutions are different, for example, the current solution is customer 8 and the target solution is customer 2 in the figure, modify the value of the target solution to the current solution. The solution individual of the current solution becomes 23524167. Start traversing from the current position until encountering the position of the customer encoding that is the same as the modified value of the current solution. For example, the fourth customer in the example is also 2. Replace the current position with the customer encoding before the current solution is modified. As shown in the figure, replace 2 with customer encoding 8. The solution individual of the current solution becomes 23584167. Then continue to compare the second position of the two solutions until the entire solution is traversed.

[0049] As Figure 4 shown, the partially matched crossover operator: First, determine two parents to be exchanged, denoted as parent 1 and parent 2. Then randomly generate two crossover points. For example, [1, 2] in the figure is to exchange the encoding of the 1-2 segment in the two parents. The offspring after the exchange may have duplicates and need to be revised. Keep the exchanged segment unchanged and find duplicate values in the non-exchanged segments. For example, both 1 and 4 in the exchanged parent 1 are duplicates with the subsequent encodings. Then find the corresponding elements at the corresponding positions in the part of the parent that has been exchanged and replace them. For example, 4 corresponds to 2 and 1 corresponds to 9. Similarly, revise the duplicate elements in parent 2.

[0050] As Figure 5 shown, the multiple 2-opt operator: Randomly select two points in the current solution sequence and swap their positions. For example, in 9354602178, swap the positions of 5 and 2, and the sequence will become 9324605178. Repeat the above steps and set the number of loops. When the number of loops reaches, jump out of the loop.

[0051] As Figure 6 shown, the random deletion and greedy insertion operation: Randomly delete n numbers from the current individual, and then traverse and insert the deleted customer numbers into the current solution individual. Take the solution with the maximum fitness as the new solution.

[0052] As Figure 7 shown, the inversion mutation operator operation: For example, the length of the solution is 10. First, randomly generate two numbers i and j in the interval [0, 9], and ensure that i < j. Assume i = 3 and j = 3. Then perform an inversion operation on the encoding sequence between positions 3 and 6 of the solution. Denote the inverted solution as the new solution.

[0053] As Figure 8 As shown, the algorithm first uses a chaotic map to generate evenly dispersed solutions within the high-dimensional solution space, enhancing the breadth of exploration. It then sequentially explores and optimizes the hippopotamuses in half the population, using the adaptive variable T to select different exploration actions. As the number of iterations increases, the algorithm dynamically adapts the initial search width and the later search speed requirements. It uses a partial matching operator to update the hippopotamus' positions, and employs a Cauchy mutation with a long-tail effect to enhance the randomness of the exploration. Next, it optimizes the remaining half of the population, generating a random new solution within the solution space, called the predator. The fitness of the current hippopotamus is compared with that of the predator. If the predator is superior, the hippopotamus' position update is discretized, and the hippopotamus moves toward the superior predator. Otherwise, the hippopotamus moves randomly. Finally, the hippopotamus enter the escape phase: the entire hippopotamus population migrates, performing local optimization to accelerate convergence. The VNS search operator employs an improved damage repair operator. Finally, an insurance mechanism to prevent local optimality is added: In order to prevent the solution from being trapped in the local optimal state for a long time, an insurance mechanism is designed. When a hippopotamus individual does not move for a long time, the present invention uses a mutation operator and applies the Metripolis criterion to make the hippopotamus move randomly out of the local optimal state to avoid premature convergence, thereby improving the convergence efficiency and solution accuracy, and reducing time waste and resource consumption in the cold chain logistics distribution process.

[0054] A cold chain logistics distribution path optimization method based on deep neural network and improved discrete Hippo algorithm, the specific steps are as follows:

[0055] A cold chain logistics distribution path optimization method based on deep neural network and improved discrete Hippo algorithm, the specific steps are as follows:

[0056] Step 1: Use natural number encoding to represent each individual, where each individual corresponds to a solution. The structure of the solution represents the delivery order sequence of customers. Assuming that the number of customers who need to unload is n, the length of the solution is n;

[0057] Step 2: Use the chaotic mapping method to generate the initial solution. This method first obtains a new solution for the population size through circle chaotic mapping, then maps it to an integer sequence with a maximum value equal to the number of customers n, and finally checks and replaces repeated numbers.

[0058] Step 3: Use a deep neural network to predict the traffic speed in the delivery area. Based on the traffic speed, the time cost, fuel consumption, flower decay cost, cooling cost, carbon emission cost, and time penalty cost are calculated by taking into account the customer's task volume and location. An optimization objective function is constructed to minimize the overall cost. The inverse of this objective function is used as the algorithm's fitness function. A larger fitness value indicates a better solution performance.

[0059] Step 4: Traverse half of the individuals in the population. During the traversal process, three different processing solutions are selected through an adaptive parameter T that changes with the number of iterations. The first one is that when the value of T is less than 0.6, the current traversal solution uses an improved discrete hippopotamus movement method to move the current solution toward the optimal solution. The principle is to compare the current solution with the optimal solution bit by bit. If the current position is different, the current solution is modified. At the same time, the current solution is traversed bit by bit to find the same code as the modified value, and it is modified to the value before the modification at the modified position. The current solution is traversed bit by bit in this way: the second one is that when the value of T is greater than 0.6, and if a randomly generated If the value of T is greater than 0.5, a partial matching crossover operation is used to move and transform the individuals in the current population; the partial matching crossover operation refers to selecting part of the code from the parent generation and directly retaining it in the offspring to generate a new solution; the third method is that when the value of T is greater than 0.6, and if a randomly generated value is less than 0.5, a non-repeating sequence from 1 to n is randomly generated to simulate the large-scale random movement of individuals in the population: then one of the methods is selected to move the individuals, and the fitness before and after the transformation is compared to decide whether to apply the new solution. According to the above rules, half of the population is traversed to complete the first stage of the improved Hippo algorithm: the exploration stage;

[0060] Step 5: Then enter the second stage of the improved hippopotamus optimization algorithm: defense stage: traverse the other half of the individuals in the population, randomly generate a new solution and compare the fitness with the current solution, simulate the hippopotamus defense predator stage in nature, when the predator's fitness is greater than the current solution, move toward the predator by improving the discrete hippopotamus movement method, simulate the natural hippopotamus taking a step forward to confront the hippopotamus, when the predator's fitness is greater than the current solution, use multiple 2-opt operations to move toward the optimal solution leading hippo in the population, and decide whether to apply the new solution by comparing the fitness before and after the transformation, and traverse the other half of the population according to the above rules.

[0061] Step 6: Then enter the third phase of the improved Hippo optimization algorithm: the defense phase: re-traverse all individuals in the population. Each traversal performs a VNS search on the current individual. The VNS search operator uses multiple random deletion greedy insertion operators;

[0062] Step 7: Based on the solutions obtained in Steps 4, 5, and 6, an insurance mechanism is added to prevent the population from falling into the local optimum. The mechanism uses a random reversal mutation operator and applies the Metripolis criterion to make the hippopotamus randomly move out of the local optimum. It detects whether the current iterative individual solution has not been updated for more than five times. If the solution has not been updated for many times, the mutation operator is used to mutate the current solution multiple times, and the Metripolis criterion is used to accept the poor solution in the hope of obtaining a better solution.

[0063] The present invention is used to solve the problems of fuel consumption, delivery time, flower loss, refrigeration cost, carbon emission cost and time penalty cost increase in the delivery process due to limited delivery vehicle resources and lack of reasonable and efficient vehicle intelligent scheduling and route planning in the cold chain logistics chain. Taking into account the customer location and needs as well as the delivery vehicle parameters, a cold chain logistics distribution intelligent scheduling route optimization model is constructed with the goal of minimizing delivery time, minimizing fuel consumption cost, flower loss, time penalty cost and additional refrigeration, carbon emission and other costs caused by unreasonable routes, and an improved discrete Hippo optimization algorithm based on real-time road condition prediction is proposed for solution.

[0064] The specific embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the scope of the present invention.

Claims

1. A cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm, characterized by: include: Flowers, goods that require delivery vehicles to complete delivery; The client, the recipient of the flowers; Refrigerated warehouses for storing flowers that need to be delivered by refrigerated vehicles; Refrigerated vehicles are used to deliver flowers to customers at designated locations and complete unloading; The intelligent dispatching system is used to dispatch refrigerated vehicles to deliver flowers from the refrigerated warehouse to the customer's designated location based on the customer's appointment.

2. The cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm according to claim 1 is characterized by: The flowers are easy to rot and wither and difficult to preserve for a long time.

3. The cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm according to claim 1 is characterized by: At the initial time of dispatch, all refrigerated vehicles are available and depart from the refrigerated warehouse location.

4. The cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm according to claim 1 is characterized by: The service time window that the customer reserves with the intelligent dispatching system must be greater than or equal to the time required for the refrigerated vehicle to unload the goods.

5. The cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm according to claim 1 is characterized by: Once the unloading operation of the refrigerated vehicle at the customer's site begins, no interruption is allowed.

6. The cold chain logistics distribution path optimization system based on deep neural network and improved discrete Hippo algorithm according to claim 1 is characterized by: Each customer may only be served once by a refrigerated vehicle.

7. A cold chain logistics distribution path optimization method based on deep neural network and improved discrete Hippo algorithm, characterized by: The specific steps are as follows: Step 1: Use natural number encoding to represent each individual, where each individual corresponds to a solution. The structure of the solution represents the delivery order sequence of customers. Assuming that the number of customers who need to unload is n, the length of the solution is n; Step 2: Use the chaotic mapping method to generate the initial solution. This method first obtains a new solution for the population size through circle chaotic mapping, then maps it to an integer sequence with a maximum value equal to the number of customers n, and finally checks and replaces repeated numbers. Step 3: Use a deep neural network to predict the traffic speed in the delivery area. Based on the traffic speed, the time cost, fuel consumption, flower decay cost, cooling cost, carbon emission cost, and time penalty cost are calculated by taking into account the customer's task volume and location. An optimization objective function is constructed to minimize the overall cost. The inverse of this objective function is used as the algorithm's fitness function. A larger fitness value indicates a better solution performance. Step 4: Traverse half of the individuals in the population. During the traversal process, three different processing solutions are selected through an adaptive parameter T that changes with the number of iterations. The first one is that when the value of T is less than 0.6, the current traversal solution uses an improved discrete hippopotamus movement method to move the current solution toward the optimal solution. The principle is to compare the current solution with the optimal solution bit by bit. If the current position is different, the current solution is modified. At the same time, the current solution is traversed bit by bit to find the same code as the modified value, and it is modified to the value before the modification at the modified position. The current solution is traversed bit by bit in this way: the second one is that when the value of T is greater than 0.6, and if a randomly generated If the value of T is greater than 0.5, a partial matching crossover operation is used to move and transform the individuals in the current population; the partial matching crossover operation refers to selecting part of the code from the parent generation and directly retaining it in the offspring to generate a new solution; the third method is that when the value of T is greater than 0.6, and if a randomly generated value is less than 0.5, a non-repeating sequence from 1 to n is randomly generated to simulate the large-scale random movement of individuals in the population: then one of the methods is selected to move the individuals, and the fitness before and after the transformation is compared to decide whether to apply the new solution. According to the above rules, half of the population is traversed to complete the first stage of the improved Hippo algorithm: the exploration stage; Step 5: Then enter the second stage of the improved hippo optimization algorithm: defense stage: traverse the other half of the individuals in the population, randomly generate a new solution and compare the fitness with the current solution, simulate the hippo defense predator stage in nature, when the predator's fitness is greater than the current solution, move toward the predator by improving the discrete hippo movement method, simulate the natural hippopotamus taking a step forward to confront the hippopotamus, when the predator's fitness is greater than the current solution, use multiple 2-opt operations to move toward the optimal solution leading hippo in the population, and decide whether to apply the new solution by comparing the fitness before and after the transformation, and traverse the other half of the population according to the above rules; Step 6: Then enter the third stage of the improved Hippo optimization algorithm: defense stage: re-traverse all individuals in the population; each traversal performs VNS search on the current individual, and the VNS search operator uses multiple random deletion greedy insertion operators; Step 7: Based on the solutions obtained in Steps 4, 5, and 6, an insurance mechanism is added to prevent the population from falling into the local optimum. The mechanism uses a random reversal mutation operator and applies the Metripolis criterion to make the hippopotamus randomly move out of the local optimum. It detects whether the current iterative individual solution has not been updated for more than five times. If the solution has not been updated for many times, the mutation operator is used to mutate the current solution multiple times, and the Metripolis criterion is used to accept the poor solution in the hope of obtaining a better solution.

Images