A method for home medical path and scheduling of a drone in cooperation with a caregiver

Abstract

The present application relates to the technical field of home medical path optimization, in particular to a home medical path and scheduling method of unmanned aerial vehicle and nursing staff cooperation, through the fusion of iterative local search algorithm and the decision-making ability of deep Q network in complex state, a hybrid intelligent optimization system is constructed, multi-objective collaborative optimization of operation cost and weight penalty caused by violation of time window is realized, and it is suitable for complex home medical service scenarios with multiple requirements such as skill matching, time window and maximum workload.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare and intelligent optimization technology, specifically to a home healthcare route and scheduling method based on multi-objective optimization and collaboration between drones and caregivers, which can be applied to scenarios such as home-based medical services, remote medical support, and collaborative scheduling of medical logistics. Background Technology

[0002] With the accelerating development of an aging society and the continuous increase in the number of patients with chronic diseases, the demand for home-based medical services is growing rapidly. Home-based medical care typically involves caregivers providing door-to-door services to clients scattered across different geographical locations. Simultaneously, some services (such as blood tests and sample collection) require the timely return of medical samples to a medical center for processing. Traditional methods mainly employ mathematical programming models (such as mixed-integer linear programming) or metaheuristic algorithms for single-objective optimization, but these methods have significant limitations in handling this complex problem. Existing technologies face several technical bottlenecks:

[0003] First, while traditional mathematical programming methods (such as CPLEX modeling) can guarantee the optimality of the solution, the model solution time increases exponentially when faced with complex conditions such as skill level constraints, time windows, and maximum workload, making it difficult to meet scheduling requirements. CPLEX models contain multiple types of decision variables and multiple sets of constraints, and in practical applications, they often fall into the "curse of dimensionality" due to computational complexity.

[0004] Secondly, existing algorithms suffer from premature convergence and a single search direction when solving this problem. Although tournament selection and ranking can maintain population diversity, conventional crossover and mutation operators are difficult to effectively explore non-dominated solution regions in the solution space, especially when dealing with discrete-continuous mixed variables.

[0005] Third, existing home-based medical care scheduling methods lack effective domain knowledge guidance mechanisms. Traditional random mutation (PM operator) and arithmetic crossover operations struggle to utilize experiential information accumulated during historical scheduling processes, such as customer time window distribution characteristics, caregiver path optimization patterns, and drone-based collaborative transportation models, resulting in limited solution quality and search efficiency. Although some studies have attempted to introduce intelligent mechanisms to improve search capabilities, an adaptive search strategy based on reinforcement learning has not yet been developed. This makes it difficult to dynamically select appropriate neighborhood operations based on the optimization process, thus limiting further improvements in multi-objective optimization performance.

[0006] Therefore, it is necessary to propose a home-based medical care scheduling framework that integrates mathematical modeling methods, reinforcement learning methods, and intelligent optimization mechanisms to achieve optimized allocation of caregiver routes and drone-assisted transportation tasks. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and to design a home medical care path and scheduling method that integrates drones and caregivers. By combining the intelligent reasoning capabilities of improved iterative local search and deep Q-networks, collaborative optimization is achieved.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A method for coordinating home healthcare delivery using drones and caregivers, characterized by the following steps: S1. Construct a chromosome coding structure for home-based medical services: uniformly encode the customer access order, caregiver allocation relationship and drone mission to form an individual solution; S2. Heuristic rule generation of initial population: The initial population is generated based on the skill matching between clients and caregivers, the client service time window, and operating costs. S3, Destruction and Reconstruction: Randomly change the access order of a caregiver path in a chromosome and accept a new solution with a certain probability to ensure the diversity of the population.

[0009] S4. Perform multi-constraint repair: Perform constraint correction on the generated solution to ensure that it meets the customer unique service constraint, skill matching constraint, time window constraint and caregiver workload constraint. S5. Construct a multi-objective optimization evaluation system: Establish a multi-objective optimization evaluation system that considers both operating costs and weight penalties for violating time windows. S6. Design the algorithm framework: Design an iterative local search based on deep Q-network: Select a neighborhood search strategy through deep reinforcement learning, iteratively optimize the population, and update the population by combining non-dominated sorting and crowding distance. S7. Iteratively solve the problem instance based on the set parameters, record the optimization process and output the Pareto non-dominated solution set, and evaluate the algorithm performance at the same time; S8. Results Storage and Visualization: Establish a results storage and visualization system to persistently store, analyze in multiple dimensions, and intuitively display optimization results.

[0010] As a further technical solution of the present invention, each individual chromosome in step S1 is a chromosome containing A doubly linked integer string of tuples, i.e. ,in This indicates the number of caregivers. For each element... , It is an integer substring containing a set of customers, whose relative order represents the service order. Furthermore, This is for The coding process for caregivers serving customers in China is as follows: First, based on the number of caregivers... To determine the chromosomes are composed of The system consists of several basic units, each corresponding to a nursing pathway. Next, all clients are divided into several subsets and assigned to different caregivers. For each caregiver, the clients they are responsible for are arranged in the order of their visits, forming an ordered client sequence. Then, the caregiver identification was applied. With the corresponding customer sequence Combine to form a binary tuple Finally, all pairs are arranged in a predetermined order (such as caregiver number order) to form a complete chromosome representation. Through this encoding method, the multi-caregiver path planning and client allocation problem can be uniformly represented as a clearly structured discrete string. Simultaneously, this encoding method directly reflects the client service order and implicitly includes caregiver allocation relationships, facilitating crossover and mutation operations in genetic algorithms. During the decoding process, for each pair... The caregivers departed from the medical center, following the sequence. The service is provided to the client in sequence. When a sample client is visited, a drone delivery task is triggered immediately after the service is completed. The drone will send the sample back to the medical center and fly to the next client the caregiver is serving to meet up, thus forming a feasible collaborative scheduling scheme between caregivers and drones.

[0011] As a further technical solution of the present invention, the specific process of step S2 is as follows: S21. Generate an initial solution based on skill matching rules: The closer the skill requirements of caregivers and clients are, the better it is to reduce service costs. Specifically: 1) Sort all clients in descending order of their skill requirements; 2) Considering the maximum number of clients a caregiver can serve, divide all clients into categories accordingly. Group, among which For the number of customers, 3) Assign as many caregivers as possible to groups as possible with skills similar to those required by the clients; S22. Generating an initial solution based on time window rules: This aims to consider the customer's time window factor and reduce weighted penalties. 1) Sort all customers in ascending order based on the median of their time windows; 2) Construct a solution without any customers. Group; 3) Assign the first customer to Group 1, the second customer to Group 2, and so on, until the [number missing]th customer. The customer was assigned to the first... Group. Then the first Each client is assigned to Group 1, and this cyclical assignment process continues until all clients have been assigned to groups; 4) Assign caregivers with skill levels close to the client's requirements to each group as much as possible; S23. Generate an initial solution based on the operating cost rule: This rule aims to reduce operating costs by simultaneously considering service costs and travel costs. 1) Use the provided c caregivers to plan c routes from the healthcare center; 2) Randomly sort all customers; 3) Assign the first customer to the route with the lowest operating cost, then delete the customer; 4) Repeat step 3) until all customers have been assigned. S24. Combine random generation strategies to supplement population diversity.

[0012] As a further technical solution of the present invention, the specific process of step S3 is as follows: S31. Randomly select a caregiver path; S32. Randomly shuffle the order of customers in the path; S33. Path reconstruction based on UAV sample transportation constraints; S34. Based on the non-dominant relationship between the new and old solutions and the probability criterion, decide whether to accept the new solution: 1) If the new solution dominates the old solution, then accept the new solution; 2) If the old solution dominates the new solution, then retain the old solution and reject the new solution; 3) If the new solution and the old solution are complementary, then accept the new solution with a probability of 0.5.

[0013] As a further technical solution of the present invention, the specific process of step S4 is as follows: S41. Skills Matching and Restoration: Ensure that caregivers' skills are not lower than the client's needs; S42, Time Window Repair: Adjust service times to reduce lateness or early arrival; S43. Service Uniqueness Restoration: Ensure that each customer is served only once; S44. Workload Constraint Repair: Limit the number of clients a caregiver can serve; S45. Drone Collaborative Constraint Repair: Ensure that sample customers are transported by drone in a timely manner.

[0014] As a further technical solution of the present invention, the objective function in step S5 is: Objective 1: Minimize operating costs, including caregiver service costs, transportation costs, and drone delivery costs; Objective 2: Minimize the weighted penalty for violating the customer's time window.

[0015] As a further technical solution of the present invention, the specific process of step S6 is as follows: S61. Treat individuals in a population as intelligent agents; S62. Construct state characteristics, including cost, penalty, and their statistical properties; S63. Define neighborhood search operators as a set of actions; S64, Adopt - Greedy strategy for selecting search operators; S65. Design a reward function based on the improved solution: Let... and These are the old and new solutions, respectively. This represents the reward value. The reward function is defined as follows: 1) If Dominate ,but Set to 5; 2) If quilt Domination, then Equal to -4; 3) If and If they do not control each other, then Set to 3; d) If and If they are the same, then Equal to 0; S66. Adaptive search is achieved using a deep Q-network update strategy. In this network, the input layer has 8 nodes, each corresponding to a state feature. The network has 16 hidden layers, each with 32 nodes, and the activation function is ReLU.

[0016] As a further technical solution of the present invention, the specific process of step S7 is as follows: S71, Loading Problem Instances: Loads instance data for home healthcare route scheduling problems, including parameters such as the number of customers, the number of caregivers, the number of drones, customer time windows, service time, and skill requirements; S72. Algorithm Parameter Configuration: Configure the algorithm running parameters, including population size, crossover probability, mutation probability, and parameters related to the deep Q-network. These parameters can be adjusted according to the problem size. S73. Iterative optimization execution: Execute an iterative optimization process based on evolutionary mechanisms and deep reinforcement learning. Each generation includes individual destruction and reconstruction, selection of neighborhood search methods by deep Q-network, target evaluation and non-dominated ranking, until the termination condition is met. S74. Results Recording and Analysis: Record the objective function values, constraint satisfaction, and Pareto front changes of each generation of the population for subsequent analysis and visualization. S75, Pareto Front Output: Outputs the final non-dominated solution set, forming the Pareto front, providing multiple scheduling schemes for the trade-off between operating costs and time window penalties; S76. Performance Evaluation and Verification: The performance of the algorithm is evaluated using the hypervolume index and the IGD index to verify the convergence of the algorithm and the distribution quality of the solution.

[0017] As a further technical solution of the present invention, the specific process of step S8 is as follows: S81. Results File Organization: A hierarchical file organization structure is adopted, with results files organized according to problem type, instance number, and number of runs. Each run instance contains multiple files including population data, Pareto frontier, and hypervolume index. S82. Data Format Standardization: The optimization results are stored using standard data formats, including numerical data in .dat format and visualization charts in .png format. The data format supports cross-platform compatibility and subsequent analysis and processing. S83, Pareto Front Visualization: Generates a scatter plot of the Pareto front, which intuitively shows the trade-off between total time and total energy consumption. The chart includes axis labels, legend, and title information to facilitate the interpretation of the results. S84. Convergence Process Analysis: Record and visualize the convergence process of the algorithm, including the trend of the objective function value with the number of generations and the population diversity index; S85. Comparison of Multiple Run Results: Comparative analysis of the results of multiple independent runs, evaluating the stability and reliability of the algorithm through statistical indicators, including mean, standard deviation, and confidence interval; S86. Result Export and Sharing: Export and share result data in multiple formats.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. Regarding algorithm architecture and adaptive search, this invention proposes an iterative local search algorithm driven by a deep Q-network. This algorithm can perceive the current solution's state characteristics (such as customer distribution and constraint violation degree) in real time and dynamically select the optimal neighborhood search operator through a reinforcement learning mechanism. Compared to traditional fixed operators or random strategies, this mechanism can adaptively balance the algorithm's exploration and development capabilities, effectively avoiding the search process from getting trapped in local optima and improving the efficiency of solving complex path scheduling problems.

[0019] In terms of domain knowledge integration and optimization quality, this invention integrates a local search strategy based on problem knowledge. By introducing problem properties specific to the characteristics of home healthcare pathways, the search space is updated accordingly. This enables the algorithm to maintain a very high proportion of feasible solutions and Pareto front convergence even when dealing with large-scale instances with strong constraints such as skill matching, time windows, and maximum workload. It also demonstrates excellent performance in minimizing operating costs and reducing weight penalties.

[0020] 3. In terms of decision support and scenario adaptability, this invention constructs a refined multi-objective evaluation system that can simultaneously optimize operating costs and weighted penalties. This system can flexibly respond to various constraint conflicts in dynamic environments, automatically correct illegal paths, and provide home-based medical institutions with diversified scheduling solutions that balance cost-effectiveness and customer satisfaction. Attached Figure Description

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments, but the scope of the invention is not limited in any way.

[0022] Figure 1 This is a schematic block diagram illustrating the workflow of the present invention.

[0023] Figure 2 This is a schematic diagram illustrating an example of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the embodiments and accompanying drawings.

[0025] In this embodiment, The parameter settings are as follows: It is a collection of customers. , It is an integer. For the number of customers; It is a set of nodes. 0 represents the medical center; customers who need to have samples collected are called sample customers. It is a customer set for which samples need to be collected. ; It represents a combination of caregivers and drones. , For the number of caregivers and drones; It is a collection of caregiver skill levels. , The highest skill level; Represents a node To the node Travel time; Represents a node To the node The cost of travel per unit time; For customers By caregiver The service time required to provide the service; The cost of service per unit of time; For customers By caregiver The cost of service per unit time during service; For caregivers Skill level; For customers Skill requirements; For customers The earliest required start time for service; For customers The latest required start time for service; The maximum number of clients a caregiver can serve; It is a very large constant.

[0026] The decision variables are as follows: : As a caregiver After serving the customer Then serve customers immediately hour, ,otherwise, ; When customers By caregiver During service, ,otherwise, ; When drones The driving route is from the customer arrive hour, ,otherwise, ; When customers By drone During service, ,otherwise, .

[0027] The auxiliary variables are as follows: For caregivers Arrival at the customer Time; For caregivers Leave the customer Time; For drones Arrival at the customer Time; For drones Leave the customer The time.

[0028] The objective function is as follows: Objective function 1: Minimize operating costs. Objective function 2: Minimize the time penalty caused by violating the time window. ; The constraint is as follows: Each customer can only be accessed once. , Flow balance: Each caregiver may only leave and return to the medical center once. , The number of caregivers leaving the medical center was equal to the number returning to the medical center. Caregivers cannot go directly from one client's home to another. Each client may only receive services from one caregiver. Relationships between decision variables: The skill level of caregivers should not be lower than the skill requirements of the clients they serve. The time when the caregiver arrives at the first client they are visiting: The time the caregiver leaves the client: The time it takes for the caregiver to reach the remaining clients they are visiting: Each customer can only receive access from one drone. Each customer can only receive access from a maximum of one drone. , Relationships between decision variables: , , The number of drones entering the medical center should equal the number of drones leaving. The drone should leave the client no earlier than the caregiver leaves the client. The time it takes for the drone to arrive at a customer's location: , Each caregiver may not serve more than the maximum number of clients that a caregiver is permitted to serve. Actual service cost per unit time: .

[0029] like Figure 1 and Figure 2 As shown, this embodiment provides a home medical care route and scheduling method for collaborative transportation of drones and caregivers, solving the scheduling optimization problem of collaborative transportation of drones and caregivers. Each customer needs to provide their expected service time window and skill requirements, while also considering the maximum workload constraint of the caregiver. The method includes the following steps: S1. Constructing a chromosome encoding structure for home-based medical services: A double-chained integer string encoding method is used to uniformly represent caregiver paths and customer assignments, resulting in the encoding structure of the parent chromosome. In this embodiment, each individual consists of several tuples, where each tuple corresponds to a caregiver service path. Specifically, the chromosome is implicitly expressed by two parts of information: first, the customer access order, which represents the service priority relationship through the arrangement of customers in the substring; second, the caregiver assignment relationship, where each substring corresponds to a caregiver, and the set of customers it contains represents the set of customers served by that caregiver.

[0030] A chromosome consists of two parts: a caregiver code and a client code. The client code contains... discrete variables ( The customer code represents the number of customers, and their relative order indicates the order in which caregivers visit them. The customer code ranges from [0, 1] to [0]. Discrete variables between [-1].

[0031] The coding process is as follows: First, based on the total number of caregivers, individual codes are determined by sequentially arranging the same number of basic units. Each basic unit corresponds to one caregiver and contains two parts of information: the first part is the caregiver's ID number, and the second part is the client access sequence assigned to that caregiver.

[0032] In the specific construction process, firstly, the set of clients each caregiver is responsible for serving is determined, and these clients are arranged into an ordered sequence according to the order in which they are served. Then, the caregiver's ID number is combined with the client sequence to form a unit. All caregiver-related units are then concatenated according to the caregiver ID order or a predetermined order to form a complete doubly linked integer string. In this string, each caregiver ID is directly identified by an integer, and the client sequence consists of client IDs listed sequentially in integer form.

[0033] This embodiment provides a chromosome encoding example: Assuming there are 9 clients, 3 caregivers, and 3 drones, the encoding is: [Caregiver allocation (3) | Service order (9)], Caregiver 1: [0,1,0,2,3,0,4,0] Caregiver 2: [0,9,0,7,0] Caregiver 3: [0,5,0,8,6,0] Decoded scheduling scheme: Caregiver 1 departs from the medical center to visit clients 1, 2, 3, and 4 in sequence and returns to the medical center. After the caregiver has visited clients 1 and 3, the drone returns to the medical center to deliver samples and then returns to meet up with the caregiver.

[0034] Caregiver 2 departs from the medical center to visit clients 9 and 7 in sequence and returns to the medical center. After the caregiver finishes visiting client 9 and returns to the medical center to deliver the sample, the drone returns to meet up with the caregiver.

[0035] Caregiver 3 departs from the medical center to visit clients 5, 8, and 6 in sequence and returns to the medical center. After the caregiver finishes visiting client 5 and returns to the medical center to deliver the sample, the drone returns to meet up with the caregiver.

[0036] S2. Heuristic rule generation of the initial population: An initial population is generated based on the matching of client and caregiver skills, the client service time window, and operating costs. The specific process is as follows: S21. Generate an initial solution based on skill matching rules: The closer the skill requirements of caregivers and clients are, the better it is to reduce service costs. Specifically: 1) Sort all clients in descending order of their skill requirements; 2) Considering the maximum number of clients a caregiver can serve, divide all clients into categories accordingly. Group, among which For the number of customers, 3) Assign as many caregivers as possible to groups as possible with skills similar to those required by the clients; S22. Generating an initial solution based on time window rules: This aims to consider the customer's time window factor and reduce weighted penalties. 1) Sort all customers in ascending order based on the median of their time windows; 2) Construct a solution without any customers. Group; 3) Assign the first customer to Group 1, the second customer to Group 2, and so on, until the [number missing]th customer. The customer was assigned to the first... Group. Then the first Each client is assigned to Group 1, and this cyclical assignment process continues until all clients have been assigned to groups; 4) Assign caregivers with skill levels close to the client's requirements to each group as much as possible; S23. Generate an initial solution based on the operating cost rule: This rule aims to reduce operating costs by considering both service costs and travel costs. 1) Use the provided c caregivers to plan c routes from the health care center; 2) Randomly sort all customers; 3) Assign the first customer to the route with the lowest operating cost, and then delete the customer; 4) Repeat step 3) until all customers have been assigned. S24. Combine random generation strategies to supplement population diversity.

[0037] The heuristic initialization method provided in this embodiment improves the quality of the initial population, significantly enhancing the performance of the algorithm and the quality of the solution.

[0038] S3, Disruption and Reconstruction: Randomly change the visit order of a caregiver path on a chromosome and accept a new solution with a certain probability, ensuring population diversity. The specific process is as follows: S31. Randomly select a caregiver path; S32. Randomly shuffle the order of customers in the path; S33. Path reconstruction based on UAV sample transportation constraints; S34. Based on the non-dominant relationship between the new and old solutions and the probability criterion, decide whether to accept the new solution: 1) If the new solution dominates the old solution, then accept the new solution; 2) If the old solution dominates the new solution, then retain the old solution and reject the new solution; 3) If the new solution and the old solution are complementary, then accept the new solution with a probability of 0.5.

[0039] The destruction and reconstruction mechanism provided in this embodiment ensures population diversity and avoids the algorithm getting trapped in local optima.

[0040] S4. Perform multi-constraint repair: Correct the constraints of the generated solution to ensure it meets the customer unique service constraint, skill matching constraint, time window constraint, and caregiver workload constraint; the specific process is as follows: S41. Skills Matching and Restoration: Ensure that caregivers' skills are not lower than the client's needs; S42, Time Window Repair: Adjust service times to reduce lateness or early arrival; S43. Service Uniqueness Restoration: Ensure that each customer is served only once; S44. Workload Constraint Repair: Limit the number of clients a caregiver can serve; S45. Drone Collaborative Constraint Repair: Ensure that sample customers are transported by drone in a timely manner.

[0041] The multi-objective optimization evaluation system provided in this embodiment offers a clear optimization direction for genetic algorithms and supports the search for Pareto optimal solutions.

[0042] S5. Construct a multi-objective optimization evaluation system: Establish a multi-objective optimization evaluation system that simultaneously considers two objectives: operating costs and weight penalties due to violations of time windows; the specific objective function is as follows: Objective 1: Minimize operating costs, including caregiver service costs, transportation costs, and drone delivery costs; Objective 2: Minimize the weighted penalty for violating the customer's time window.

[0043] S6. Algorithm Framework Design: An iterative local search based on a deep Q-network is designed. This involves selecting a neighborhood search strategy through deep reinforcement learning, iteratively optimizing the population, and updating the population by combining non-dominated ranking and crowding distance. The algorithm supports multiple independent runs to ensure the stability and reliability of the results. The specific process is as follows: S61. Treat individuals in a population as intelligent agents; S62. Construct state characteristics, including cost, penalty, and their statistical properties; S63. Define neighborhood search operators as a set of actions; S64, Adopt - Greedy strategy for selecting search operators; S65. Design a reward function based on the improved solution: Let... and These are the old and new solutions, respectively. This represents the reward value. The reward function is defined as follows: 1) If Dominate ,but Set to 5; 2) If quilt Domination, then Equal to -4; 3) If and If they do not control each other, then Set to 3; d) If and If they are the same, then Equal to 0; S66. Adaptive search is achieved using a deep Q-network update strategy. In this network, the input layer has 8 nodes, each corresponding to a state feature. The network has 16 hidden layers, each with 32 nodes, and the activation function is ReLU.

[0044] S7. Iteratively solve the problem instance based on the set parameters, record the optimization process and output the Pareto non-dominated solution set, and evaluate the algorithm performance; the specific process is as follows: S71, Loading Problem Instances: Loads instance data for home healthcare route scheduling problems, including parameters such as the number of customers, the number of caregivers, the number of drones, customer time windows, service time, and skill requirements; S72. Algorithm Parameter Configuration: Configure the algorithm running parameters, including population size, crossover probability, mutation probability, and parameters related to the deep Q-network. These parameters can be adjusted according to the problem size. S73. Iterative optimization execution: Execute an iterative optimization process based on evolutionary mechanisms and deep reinforcement learning. Each generation includes individual destruction and reconstruction, selection of neighborhood search methods by deep Q-network, target evaluation and non-dominated ranking, until the termination condition is met. S74. Results Recording and Analysis: Record the objective function values, constraint satisfaction, and Pareto front changes of each generation of the population for subsequent analysis and visualization. S75, Pareto Front Output: Outputs the final non-dominated solution set, forming the Pareto front, providing multiple scheduling schemes for the trade-off between operating costs and time window penalties; S76. Performance Evaluation and Verification: The performance of the algorithm is evaluated using the hypervolume index and the IGD index to verify the convergence of the algorithm and the distribution quality of the solution.

[0045] S8. Results Storage and Visualization: Establish a results storage and visualization system to persistently store, multi-dimensionally analyze, and intuitively display optimization results, providing strong support for performance evaluation and decision support. The specific process is as follows: S81. Results File Organization: A hierarchical file organization structure is adopted, with results files organized according to problem type, instance number, and number of runs. Each run instance contains multiple files including population data, Pareto frontier, and hypervolume index. S82. Data Format Standardization: The optimization results are stored using standard data formats, including numerical data in .dat format and visualization charts in .png format. The data format supports cross-platform compatibility and subsequent analysis and processing. S83, Pareto Front Visualization: Generates a scatter plot of the Pareto front, which intuitively shows the trade-off between total time and total energy consumption. The chart includes axis labels, legend, and title information to facilitate the interpretation of the results. S84. Convergence Process Analysis: Record and visualize the convergence process of the algorithm, including the trend of the objective function value with the number of generations and the population diversity index; S85. Comparison of Multiple Run Results: Comparative analysis of the results of multiple independent runs, evaluating the stability and reliability of the algorithm through statistical indicators, including mean, standard deviation, and confidence interval; S86. Result Export and Sharing: Export and share result data in multiple formats.

[0046] This embodiment addresses the multi-objective home healthcare routing and scheduling problem considering drone assistance, constructing a deep reinforcement learning-driven multi-objective optimization algorithm framework to achieve efficient and intelligent solutions to complex service scheduling problems. This method is based on iterative local search and introduces a deep Q-network to adaptively control the search strategy, thereby achieving a synergistic improvement in search efficiency and solution quality during the multi-objective optimization process.

[0047] This example focuses on deep reinforcement learning-driven iterative local search, improving upon traditional iterative local search algorithms. The main differences compared to standard iterative local search are as follows: First, a deep Q-network is introduced to select a suitable neighborhood selection method, enabling adaptive selection of local search strategies and thus improving search efficiency and solution quality. Second, a chromosome decoding and constraint repair mechanism is constructed to perform multi-stage feasibility adjustments on the generated solutions to meet constraints such as time windows, skill matching, and resource allocation. Third, the individual selection mechanism is improved to be compatible with solutions generated based on reinforcement learning and to support comparison strategies based on Pareto dominance and non-dominated ranking levels. Furthermore, to enhance algorithm stability, a perturbation mechanism and acceptance criteria are combined during the search process to achieve effective escape from local optima.

[0048] Algorithms and processes not described in detail in this article are general techniques in this field.

[0049] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0050] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A problem concerning the collaborative transportation of home-based medical care using drones and caregivers, characterized in that, Includes the following steps: S1. Construct a chromosome coding structure for home-based medical services: uniformly encode the customer access order, caregiver allocation relationship and drone mission to form an individual solution; S2. Heuristic rule generation of initial population: The initial population is generated based on the skill matching between clients and caregivers, the client service time window, and operating costs. S3, Destruction and Reconstruction: Randomly change the access order of a caregiver path in a chromosome and accept a new solution with a certain probability to ensure the diversity of the population; S4. Perform multi-constraint repair: Perform constraint correction on the generated solution to ensure that it meets the customer unique service constraint, skill matching constraint, time window constraint and caregiver workload constraint. S5. Construct a multi-objective optimization evaluation system: Establish a multi-objective optimization evaluation system that considers both operating costs and weight penalties for violating time windows. S6. Design the algorithm framework: Design an iterative local search based on deep Q-network: Select a neighborhood search strategy through deep reinforcement learning, iteratively optimize the population, and update the population by combining non-dominated sorting and crowding distance. S7. Iteratively solve the problem instance based on the set parameters, record the optimization process and output the Pareto non-dominated solution set, and evaluate the algorithm performance at the same time; S8. Results Storage and Visualization: Establish a results storage and visualization system to persistently store, analyze in multiple dimensions, and intuitively display optimization results.

2. As a further technical solution of the present invention, each individual chromosome in step S1 is a single chromosome containing... A doubly linked integer string of tuples, i.e. ,in This indicates the number of caregivers. For each element... , It is an integer substring containing a set of customers, whose relative order represents the service order. This is for The coding process for caregivers serving customers in China is as follows: First, based on the number of caregivers... To determine the chromosomes are composed of The system consists of several basic units, each corresponding to a nursing pathway. Next, all clients are divided into several subsets and assigned to different caregivers. For each caregiver, the clients they are responsible for are arranged in the order of their visits, forming an ordered client sequence. Subsequently, the caregiver identification was applied. With the corresponding customer sequence Combine to form a binary tuple Finally, all the pairs are arranged in a predetermined order to form a complete chromosome representation; during the decoding process, for each pair... The caregivers departed from the medical center, following the sequence. The service is provided to the client in sequence. When a sample client is visited, a drone delivery task is triggered immediately after the service is completed. The drone will send the sample back to the medical center and fly to the next client the caregiver is serving to meet up, thus forming a feasible caregiver-drone collaborative scheduling scheme.

3. The method for coordinating home medical care routes and scheduling using drones and caregivers according to claim 1, characterized in that, The specific process of step S2 is as follows: S21. Generate an initial solution based on skill matching rules: 1) Sort all clients in descending order of their skill requirements; 2) Considering the maximum number of clients a caregiver can serve, sequentially divide all clients into... Group, among which For the number of customers, 3) Assign as many caregivers as possible to groups as possible with skills similar to those required by the clients; S22. Generate an initial solution based on time window rules: 1) Sort all customers in ascending order according to the median value of their time windows; 2) Construct a solution without any customers. Group; 3) Assign the first customer to Group 1, the second customer to Group 2, and so on, until the [number missing]th customer. The customer was assigned to the first... Group. Then the first Each client is assigned to Group 1, and this cyclical assignment continues until all clients have been assigned to groups; 4) Assign caregivers whose skill level is close to the client's requirements to each group as much as possible. S23. Generate an initial solution based on the operating cost rule: 1) Use the provided c caregivers to plan c routes from the health care center; 2) Randomly sort all customers; 3) Assign the first customer to the route with the lowest operating cost, and then delete the customer; 4) Repeat step 3) until all customers have been assigned. S24. Combine random generation strategies to supplement population diversity.

4. The home medical care route and scheduling method for collaborative transportation by drones and caregivers according to claim 1, characterized in that, The specific process of step S3 is as follows: S31. Randomly select a caregiver path; S32. Randomly shuffle the order of customers in the path; S33. Path reconstruction based on UAV sample transportation constraints; S34. Based on the non-dominant relationship between the new and old solutions and the probability criterion, decide whether to accept the new solution: 1) If the new solution dominates the old solution, then accept the new solution; 2) If the old solution dominates the new solution, then retain the old solution and reject the new solution; 3) If the new solution and the old solution are complementary, then accept the new solution with a probability of 0.

5.

5. The home medical route and scheduling method for collaborative transportation by drones and caregivers according to claim 1, characterized in that, The specific process of step S4 is as follows: S41. Skills Matching and Restoration: Ensure that caregivers' skills are not lower than the client's needs; S42, Time Window Repair: Adjust service times to reduce lateness or early arrival; S43. Service Uniqueness Restoration: Ensure that each customer is served only once; S44. Workload Constraint Repair: Limit the number of clients a caregiver can serve; S45. Drone Collaborative Constraint Repair: Ensure that sample customers are transported by drone in a timely manner.

6. The home medical care route and scheduling method for collaborative transportation by drones and caregivers according to claim 1, characterized in that, The objective function described in S5 is: Objective 1: Minimize operating costs, including caregiver service costs, transportation costs, and drone delivery costs; Objective 2: Minimize the weighted penalty for violating the customer's time window.

7. The method for coordinating home medical care routes and scheduling using drones and caregivers according to claim 1, characterized in that, The specific process of step S6 is as follows: S61. Treat individuals in a population as intelligent agents; S62. Construct state characteristics, including cost, penalty, and their statistical properties; S63. Define neighborhood search operators as a set of actions; S64, Adopt - Greedy strategy for selecting search operators; S65. Design a reward function based on the improved solution: Let... and These are the old and new solutions, respectively. This represents the reward value. The reward function is defined as follows: 1) If Dominate ,but Set to 5; 2) If quilt Domination, then Equal to -4; 3) If and If they do not control each other, then Set to 3; d) If and If they are the same, then Equal to 0; S66. Adaptive search is achieved using a deep Q-network update strategy. In this network, the input layer has 8 nodes, each corresponding to a state feature. The network has 16 hidden layers, each with 32 nodes, and the activation function is ReLU.

8. The home medical care route and scheduling method for collaborative transportation by drones and caregivers according to claim 1, characterized in that, The specific process of step S7 is as follows: S71, Problem Instance Loading: Load instance data for home healthcare route scheduling problems, including the number of customers, the number of caregivers, the number of drones, customer time windows, service time, and skill requirement parameters; S72. Algorithm Parameter Configuration: Configure the algorithm running parameters, including population size, crossover probability, mutation probability, and parameters related to the deep Q-network. These parameters can be adjusted according to the problem size. S73. Iterative optimization execution: Execute an iterative optimization process based on evolutionary mechanisms and deep reinforcement learning. Each generation includes individual destruction and reconstruction, selection of neighborhood search methods by deep Q-network, target evaluation and non-dominated ranking, until the termination condition is met. S74. Results Recording and Analysis: Record the objective function values, constraint satisfaction, and Pareto front changes of each generation of the population for subsequent analysis and visualization. S75, Pareto Front Output: Outputs the final non-dominated solution set, forming the Pareto front, providing multiple scheduling schemes for the trade-off between operating costs and time window penalties; S76. Performance Evaluation and Verification: The performance of the algorithm is evaluated using the hypervolume index and the IGD index to verify the convergence of the algorithm and the distribution quality of the solution.

9. The home medical care route and scheduling method for collaborative transportation by drones and caregivers according to claim 1, characterized in that, The specific process of step S8 is as follows: S81. Results File Organization: A hierarchical file organization structure is adopted, with results files organized according to problem type, instance number, and number of runs. Each run instance contains multiple files including population data, Pareto frontier, and hypervolume index. S82. Data Format Standardization: The optimization results are stored using standard data formats, including numerical data in .dat format and visualization charts in .png format. The data format supports cross-platform compatibility and subsequent analysis and processing. S83, Pareto Front Visualization: Generates a scatter plot of the Pareto front, which intuitively shows the trade-off between total time and total energy consumption. The chart includes axis labels, legend, and title information to facilitate the interpretation of the results. S84. Convergence Process Analysis: Record and visualize the convergence process of the algorithm, including the trend of the objective function value with the number of generations and the population diversity index; S85. Comparison of Multiple Run Results: Comparative analysis of the results of multiple independent runs, evaluating the stability and reliability of the algorithm through statistical indicators, including mean, standard deviation, and confidence interval; S86. Result Export and Sharing: Export and share result data in multiple formats.

Images