Wireless resource allocation method and system for automatic bidding

Abstract

The invention belongs to the technical field of communication network and intelligent resource scheduling, and discloses a wireless resource allocation method and system for automatic bidding and tendering, a central control node and a plurality of resource agents cooperate to form a multi-agent scheduling framework, the central node is responsible for task analysis, bidding document generation and bid winning decision, and the central control node is responsible for task scheduling. A resource agent calculates bidding willingness and quotation according to a local state, realizes iterative optimization of a limited main problem and a pricing sub-problem in combination with a column generation algorithm so as to reduce calculation complexity, and completes task resource matching and dynamic bidding and tendering processes under the constraint of multi-dimensional service quality. And meanwhile, a QoS-QoE mapping and reverse optimization model based on user experience is introduced to realize experience-driven adaptive task configuration, so that the system has high efficiency, expandability and intelligent autonomy, and is suitable for a multi-access wireless network, edge computing and distributed resource collaboration scene.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of communication network resource management and intelligent scheduling, and particularly relates to a wireless resource allocation method and system for automatic bidding and tendering. BACKGROUND

[0002] With the rapid development of new communication systems such as multi-access wireless networks, Internet of Things, and edge computing, the types of wireless resources in the network are increasingly diverse, and different nodes have obvious differences in bandwidth, latency, power, and computing capacity. How to achieve efficient task allocation and scheduling in a diversified and dynamic resource environment has become a key problem in heterogeneous network management.

[0003] Wireless resource allocation methods usually revolve around the key resource dimensions of communication systems and can be mainly divided into three categories:

[0004] Frequency resource allocation: dividing limited frequency bands to different users or cells to avoid co-channel interference. Common methods include frequency division multiple access (FDMA), orthogonal frequency division multiple access (OFDMA), etc. For example, in 4G / LTE, subcarrier allocation is used to achieve multi-user parallel transmission.

[0005] Time resource allocation: dividing time slots or symbols in the time dimension and allocating them to different users. A typical application is time division multiple access (TDMA), and flexible time slot configuration in 5G, which can dynamically adjust the proportion of time resources according to the latency requirements of business.

[0006] Power resource allocation: allocating transmit power according to user channel quality and business priority. For example, in a cellular network, the power of edge users is increased to ensure communication quality, and the power of users near the base station is reduced to reduce interference.

[0007] Traditional wireless resource allocation methods mostly rely on centralized optimization solutions or heuristic algorithms. Although mixed integer programming methods can obtain the theoretically optimal solution, the computational complexity increases dramatically as the network size expands, making it difficult to meet the requirements of real-time response. Distributed methods based on greedy strategies or contract net protocols have good computational efficiency, but they have deficiencies in global optimality, stability, and scalability. Auction-based allocation mechanisms achieve dynamic matching of tasks and resources through bidding, which can improve flexibility to some extent, but the commonly used fixed combination or enumeration search method is prone to combinatorial explosion in large-scale scenarios, making it difficult to balance computational efficiency and scheduling accuracy.

[0008] Meanwhile, the existing resource allocation mechanism often optimizes only from the network resource level, ignoring the correlation between user experience quality (QoE) and service benefit, resulting in that the allocation result is difficult to balance user satisfaction and system efficiency. In summary, the current wireless resource scheduling still faces problems such as high computational complexity, lack of global optimality, and poor dynamic adaptability. Therefore, it is necessary to design an automated resource scheduling mechanism that can realize autonomous decision-making, low-latency response, and efficient allocation in a complex heterogeneous environment to improve the overall efficiency and service quality of the system.

[0009] Wireless resource allocation refers to managing, controlling and scheduling frequency, time, power, code resources and other resources in a wireless communication system through certain strategies and means, so as to make the best use of limited wireless network resources, ensure that various businesses meet the quality of service requirements, and improve system capacity and resource utilization. Common wireless resource allocation algorithms include static allocation algorithms and dynamic allocation algorithms, and dynamic allocation algorithms include polling scheduling algorithms, maximum carrier-to-interference ratio scheduling algorithms, proportional fair scheduling algorithms, etc. SUMMARY

[0010] The purpose of the present application is to propose a wireless resource allocation method and system based on an automated bidding mechanism to solve the problems of insufficient real-time performance, high computational complexity and uneven resource utilization in existing wireless resource allocation methods. The method introduces an adaptive task-based bidding process and a distributed collaborative decision-making mode, enabling the resource allocation process to automatically match and dynamically optimize in multi-type node and multi-task scenarios, thereby significantly improving the overall operating efficiency of the system while ensuring service quality.

[0011] The purpose of the present application is also to introduce game theory and market mechanism ideas into the resource scheduling process, enabling each resource node to have autonomous decision-making capabilities and achieving an automated closed loop of bidding, evaluation and winning under the coordination of the central control node. At the same time, an incremental combination generation strategy is used to reduce the computational overhead in large-scale resource scheduling and avoid the combinatorial explosion problem caused by enumeration search in traditional methods.

[0012] The technical solution of the present application is an automated bidding wireless resource allocation method and system, which introduces a task-based bidding mechanism in the resource allocation process and realizes dynamic optimization and autonomous scheduling of resources through collaboration between the central control node and multiple resource agents. This method combines auction game and combinatorial optimization ideas in mechanism, which can effectively reduce the computational complexity, improve the scalability and real-time performance of the system while ensuring service quality.

[0013] The system as a whole is composed of a central control node and a plurality of resource agent nodes. The central control node is responsible for receiving task requests, generating a tender, screening candidate agents, initiating a bidding process, and completing a bid decision; each resource agent node represents a different wireless resource entity and has functions such as resource state perception, bid willingness judgment, and bid generation. The two interact through a standardized task message interface, thus forming an extensible multi-agent automated allocation framework.

[0014] At the method level, the system first converts the user's service demand or quality of experience (QoE) parameters into quantifiable quality of service (QoS) indicators, including minimum transmission bandwidth, maximum allowable delay, and minimum reliability. The central control node generates a standardized tender based on these indicators and selects qualified agent nodes through a candidate screening module. The selected agents calculate their bidding intentions based on their resource load and operating costs and form bid responses.

[0015] In the bidding process (mechanism), the central control node is based on an auction mechanism and uses an improved combination generation strategy to dynamically optimize agent combinations. Unlike traditional enumeration searches, the invention introduces an incremental idea: the initial stage contains only a small number of basic combinations, and the system dynamically generates new combinations with improvement potential based on the dual variables of the previous round of solution results and iteratively updates the bid evaluation results. This process converges to an allocation scheme close to the global optimum within a limited number of iterations, significantly reducing computational complexity.

[0016] To improve the scalability of the system, the invention also supports both centralized and distributed implementation architectures. In the centralized mode, all task analysis and combination optimization are completed at the central control node, which is suitable for small and medium-sized networks; in the distributed mode, combination generation and local decision-making are decentralized to each resource agent, and the central node is only responsible for coordination and global aggregation, effectively reducing the central computing load and improving parallel efficiency.

[0017] In addition, the invention introduces a user experience-based reverse mapping mechanism at the task modeling level. The system establishes a predictive relationship between QoS and QoE through a machine learning model and uses an inverse optimization algorithm to generate the corresponding optimal QoS parameters given a QoE target, allowing resource scheduling to adapt to user demand and implementing an experience-oriented resource allocation strategy.

[0018] Beneficial effects: Through the above structure and method design, the invention can maintain stable operating performance in a multi-task parallel, high dynamic load, and heterogeneous network environment, achieving a good balance between service quality, resource utilization, and runtime delay, and providing an efficient and scalable implementation scheme for future automated resource management of intelligent communication networks.

[0019] This invention also automates the entire process of wireless resource allocation, from task generation and bidding decisions to contract execution. The system can adaptively schedule various types of resources in complex heterogeneous environments, significantly improving resource utilization and scheduling response speed while ensuring task completion rates. It demonstrates good engineering feasibility and application prospects. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall system structure of the present invention;

[0021] Figure 2 This is a flowchart illustrating the automated bidding process of this invention.

[0022] Figure 3 This is a flowchart of the pre-bid logic.

[0023] Figure 4 The bidding decision-making process for resource agents;

[0024] Figure 5 This is a flowchart of the bid evaluation and award determination process;

[0025] Figure 6 The mapping relationship between QoE and QoS and the optimized structure. Detailed Implementation

[0026] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0027] like Figure 1 As shown, the system comprises a control unit, a standardized bidding module, and a candidate selection module; it is divided into two main functional domains: the control unit functional domain and the RAN-Agent functional domain; the overall system structure consists of a central control node and multiple resource agent nodes. The central control node is responsible for task parsing, bidding document generation, candidate selection, bid evaluation decision-making, and task contract issuance; the resource agent nodes represent various wireless resource entities and have functions such as status reporting, willingness calculation, bid generation, and task execution. The system achieves information exchange between nodes through standardized communication interfaces, thereby forming an automatically executable resource scheduling closed loop.

[0028] like Figure 2 As shown, the system operation flow of this invention includes six stages in sequence: task input, combination generation, bidding, tendering, bid evaluation, and contract execution. The system first receives external task requests from the central control node and converts the Quality of User Experience (QoE) or business requirement parameters into quantifiable Quality of Service (QoS) indicators, such as minimum transmission bandwidth, maximum allowable latency, and minimum reliability.

[0029] Subsequently, the central control node executes the following based on the task's QoS constraints and the real-time status information of each resource agent node: Figure 3The pre-bid logic of the application screens and optimizes possible resource allocation combinations to determine the selected target nodes.

[0030] In the initial stage of pre-bid, the application adopts a screening mechanism of "Dual-Pool Layout" to divide the candidate agent nodes into separate bidding pool and combined bidding pool. The system first calculates the comprehensive score value F0 of each accessible node according to formula 4 i (weight combination {w1, w2, w3} takes the value of 0.4, 0.3, 0.3); the system divides the actual boundary threshold of the dual pool according to the score value F0 of the task constraint, that is, for separate bidding, when F i >0.8*F0, the node is selected; for combined bidding, when F i >0.5*F0, the node is selected.

[0031] In the combined generation stage of pre-bid, the system needs to extract nodes from the combined bidding pool and generate the optimal combination. In order to avoid the problem of high computational complexity caused by global enumeration of all candidate combinations in traditional auction mechanism, the application introduces column generation (CG) algorithm for optimization. The core idea of column generation is to realize the dynamic expansion and incremental optimization of candidate combinations through the alternative solution of restricted master problem (RMP) and pricing sub-problem (PP), so as to approximate the global optimal solution within a limited number of iterations. The central control node first establishes a restricted master problem containing a small number of initial combinations, and solves it by linear relaxation to obtain the corresponding dual variables. Each resource agent node independently solves the local pricing sub-problem according to the dual information, calculates the possible high-yield combination and its reduced cost. When a new combination with negative reduced cost appears, it is added to the master problem for re-solution; if no improved combination appears in continuous iterations, it is considered that the system has reached convergence. This process realizes the dynamic evolution from local candidate to global optimal allocation, effectively reducing the overall computational load while maintaining the solution accuracy.

[0032] For ease of understanding, the column generation optimization process of the application can be expressed in the following formal pseudo code:

[0033]

[0034]

[0035] Pseudo code 1 centralized CG pseudo code

[0036]

[0037]

[0038] Pseudo code 2 distributed CG pseudo code

[0039] After the combination generation is completed and the standardized tender book is formed, the system enters the bidding phase. After each resource agent node (RAN) receives the tender book distributed by the central control node, it will independently make a decision based on the local resource state and combination recommendation information. As shown in Figure 4 The present application adopts a double-layer bidding system of "will-offer separation" to enhance the autonomy and stability of the system.

[0040] In the will decision phase of the first layer, the agent node calculates the local comprehensive score according to the available bandwidth, current delay, reliability and operating cost parameters according to formula 1, to determine whether to participate in the bidding. The score function is defined as:

[0041]

[0042] Wherein, B i , L i , R i , C i represent the bandwidth, delay, reliability and unit cost of the i-th agent node; the system configuration weight parameters α, β, γ, δ are respectively 0.5, 0.3, 0.2, 0.3, to balance performance and overhead. When the node score is higher than the set threshold K (the individual bidding K1 is 0.65, the combination bidding K2 is 0.60, and the adjustment bidding K3 is 0.55), the node determines to participate, and the corresponding will flag is set to participate; otherwise, it is set to refuse. For the nodes in the recommended combination, the system will correspondingly lower the threshold to encourage them to participate in the combination bidding, so as to improve the feasibility and diversity of the recommended combination.

[0043] In the offer generation phase of the second layer, only when the node is marked as participating in the will layer, the actual offer calculation process is entered. The generation of the offer comprehensively considers the local execution cost, expected income and combination synergy factor of the node. When multiple nodes bid in combination, the combination offer follows the concession model shown in formula 2

[0044]

[0045] Wherein, n is the number of combination members, and λ is the concession coefficient λ, which is 0.2. This model realizes the fair discount within the combination by weighted average of the offer, so that the overall offer maintains balance while reducing cost. After the generation of the will and the offer is completed, each agent node returns the bidding information to the central control node in the standardized response format. The system forms a complete set of bidding results in this phase, providing input basis for subsequent bidding evaluation and bid-winning determination.

[0046] The central control node starts the bidding evaluation process after collecting all the bidding responses, as shown in Figure 5The central control node unifies the bid information returned by each candidate node or combination, and performs differentiated scoring and judgment according to the transfer mode. The system supports two main transfer modes: split mode and redundant mode, which are suitable for high-bandwidth services and high-reliability services, respectively.

[0047] In the split mode, the task flow is divided into several bandwidth sub-flows and allocated to different proxy nodes for parallel transmission. At this time, the system focuses on the total bandwidth and load balancing of the combination, and the combination parameter calculation is as formula 3-1;

[0048]

[0049] That is, the bandwidth of the combination takes the sum, the delay takes the maximum value, and the reliability takes the minimum value to ensure that all sub-flows can be completed at the slowest node.

[0050] In the redundant mode, multiple nodes transmit the same data simultaneously to improve reliability, and the system focuses on stability and delay instead of bandwidth. The parameter calculation is formula 3-2:

[0051]

[0052] That is, the bandwidth of the combination takes the maximum value, the delay takes the minimum value, and the reliability is obtained by cumulative probability.

[0053] The system uses a comprehensive scoring function for unified sorting, and the single node or combination node with the highest score is awarded. The comprehensive scoring function is formula 4:

[0054]

[0055] Among them, the system scoring weight Q j represents the task satisfaction degree, S j represents the service level fulfillment score, P j w1, w2, w3 are system scoring weights, taking values of 0.4, 0.4, and 0.2 respectively. The single node or combination with the highest final score is determined as the winning bidder. The central control node generates a task contract and broadcasts the winning result, the winning node locks the resources and executes the task, and the non-winning node enters standby or adjustment state. The whole bidding and bidding decision-making process is completed within milliseconds, ensuring that the system can still achieve efficient and stable resource allocation in a complex heterogeneous environment.

[0056] If the system is deployed with a visual interface, the node state, task matching, and winning result of the bidding process can be dynamically presented through the front-end interface, making it convenient to monitor and debug the running state.

[0057] To achieve dynamic task generation based on user experience, this invention introduces the following in the task modeling stage: Figure 6 The module shown illustrates the QoE-QoS mapping and reverse optimization structure. This module establishes a mapping model between QoS and QoE through machine learning, automatically deriving the optimal QoS parameter set when a target QoE is specified, and using the result as new bidding input to achieve experience-driven resource scheduling. The system first continuously collects key QoS parameters from each resource proxy node during task execution, including bandwidth (B), latency (L), and reliability (R), and combines this with user feedback or task completion to generate an experience quality score (standard MOS)E. Using historical sample data, the system constructs a QoS-QoE mapping model, establishing a predictive relationship as shown in Equation 5.

[0058] E=f(B,L,R) (5)

[0059] Where f(·) is a nonlinear function obtained through training. Based on the model comparison results, the system uses the LightNN–GBDT hybrid regression model to achieve QoE prediction. This model combines the nonlinear expressive power of neural networks with the stability of gradient boosting trees, enabling it to meet the real-time requirements of online inference while maintaining accuracy. In the backpropagation phase, when the system obtains the target experience value E... target Then, the optimal QoS parameter combination (B) will be solved using the Bayesian optimization algorithm. * ,L * ,R * ), making Formula 6 hold f(B) * ,L * ,R * )≈E target (6)

[0060] The resulting parameter combinations are passed to the combination generation module as input for a new round of task requirements. Through this mechanism, the system can adaptively adjust bandwidth allocation, latency constraints, and reliability targets based on user experience feedback, thereby achieving a dynamic balance between QoS and QoE in different business scenarios. During task execution, the module also updates the QoE prediction results and calculates the experience deviation in real time. When the current experience value E is detected... t With target value E target The difference exceeds the threshold ΔE th When the value is 1.5, the central control node automatically triggers a new round of combination generation and scheduling process to achieve experience-driven closed-loop resource optimization.

[0061] This mechanism enables the present invention to not only focus on network performance indicators in resource allocation, but also incorporate user experience into the optimization objectives, significantly improving the system's service satisfaction and intelligent adaptive capabilities.

[0062] In the simulation experiment verification, the method is compared with the traditional mixed integer programming, the greedy algorithm and the contract net mechanism in performance. The results show that the method shortens the running time by about 40% under the premise of ensuring the quality of service, and the task satisfaction rate is increased to more than 0.7 on average, and has high efficiency and scalability.

[0063] In conclusion, the wireless resource allocation driven by tasks is realized through the automatic bidding mechanism between the central control node and the resource agent, the calculation complexity is significantly reduced through the incremental combination optimization, and the task adaptability is enhanced through the QoE driven reverse modeling. The system structure is clear, the calculation cost is low, the efficient and real-time resource scheduling and task matching can be realized in the heterogeneous network environment, and the system has a good engineering application prospect.

[0064] The above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A method for wireless resource allocation for automated tendering, characterized in that, Based on the wireless resource allocation system, the system is composed of a central control node and a plurality of resource agent nodes; the central control node is responsible for receiving wireless resource use task request, generating tender book, screening candidate agent, initiating bidding process and completing bidding decision; each resource agent node represents different wireless resource entity, has resource state perception, bidding willingness judgment and price generation function; the central control node and the resource agent node interact through the standardized task message interface between nodes, so as to form an extensible multi-agent automatic allocation execution resource scheduling framework; the central control node includes a control unit, a standardized tender module and a candidate screening module; the system first converts the user's service demand or quality of experience (QoE) parameter into quantifiable quality of service (QoS) index, including minimum transmission bandwidth, maximum allowable delay and minimum reliability; the central control node generates a standardized tender book according to these indexes, and selects the qualified agent node through the candidate screening module; the selected agent node calculates the bidding intention according to its resource load and operation cost, and forms the price response; In the bidding process (mechanism), the central control node takes the auction mechanism as the basis, adopts the improved combination generation strategy to dynamically optimize the selection combination of the agent node; the incremental column generation scheme is adopted: in the initial stage, only a small number of combined agent nodes are included, the system dynamically generates new combined agent nodes with improvement potential according to the dual variables of the previous round of solution results, and iteratively updates the evaluation results; this process can converge to a distribution scheme close to the global optimum within a limited iteration, thereby significantly reducing the computational complexity; In the resource allocation process, the task-based bidding mechanism is introduced, and the dynamic optimization and autonomous scheduling of resources are realized through the cooperation between the central control node and multiple resource agents; the mechanism combines the ideas of auction game and combination optimization, which can effectively reduce the computational complexity and improve the scalability and real-time performance of the system under the premise of ensuring the quality of service.

2. The method of claim 1, wherein, In order to improve the scalability of the system, both centralized and distributed implementation architectures are supported; in the centralized mode, all task analysis and combination optimization are completed in the central control node, which is suitable for small and medium-sized networks; in the distributed mode, combination generation and local decision are decentralized to each resource agent, and the central node is only responsible for coordination and global aggregation, thereby effectively reducing the center computing load and improving the parallel efficiency.

3. The method of claim 1, wherein the wireless resource allocation is performed in an automated bidding process. The reverse mapping mechanism based on user experience is introduced at the task modeling level; the system establishes the prediction relationship between QoS and QoE through machine learning model, and generates the corresponding optimal QoS parameters under the given QoE target by using inverse optimization algorithm, so that the resource scheduling can be adaptively adjusted according to user demand, and the experience-oriented resource allocation strategy is realized.

4. The method of claim 1, wherein the wireless resource allocation is performed in an automated bidding process. The system is divided into two functional domains: control body functional domain and wireless resource agent RAN-Agent functional domain; the overall structure of the system is composed of a central control node and multiple resource agent nodes; the central control node is responsible for task analysis, tender book generation, candidate screening, bid decision and task contract issuance; the resource agent node represents various wireless resource entities and has the functions of state reporting, willingness calculation, bid generation and task execution; the system realizes information interaction between nodes through a standardized communication interface, thereby forming a closed loop of resource scheduling that can be automatically executed.

5. The method for automated bidding and wireless resource allocation as claimed in claim 1, wherein, The system operation process includes six stages of task input, combination generation, tender, bid, bid evaluation and contract execution; the system first receives external task requests from the central control node, and converts the user experience quality (QoE) or service demand parameters into quantifiable quality of service (QoS) indicators, such as minimum transmission bandwidth, maximum allowable delay and minimum reliability; then, the central control node performs combination generation logic according to the QoS constraints of the task and the real-time state information of each resource agent node, and screens and optimizes possible resource allocation combinations; in order to avoid the problem of high computational complexity caused by global enumeration of all candidate combinations in traditional auction mechanism, the column generation (CG) algorithm is introduced for optimization; the core idea of column generation is to realize dynamic expansion and incremental optimization of candidate combinations through alternating solution of the restricted master problem (RMP) and the pricing sub-problem (PP), so as to approximate the global optimal solution within a limited number of iterations; in the initial stage, the central control node first establishes a restricted master problem containing a small number of initial combinations, and solves it by linear relaxation to obtain the corresponding dual variables; each resource agent node independently solves the local pricing sub-problem according to the dual information to calculate possible high-yield combinations and their reduced costs; when a new combination with negative reduced cost appears, it is added to the master problem for re-solution; If no improved combination appears in continuous iterations, the system is considered to have reached convergence.

6. The wireless resource allocation method of automatic tender and bid according to claim 1, characterized in that, After the combination generation is completed and the standardized tender book is formed, the system enters the bid stage; after each resource agent node (RAN) receives the tender book distributed by the central control node, it will independently make decisions based on local resource state and combination recommendation information; a two-level bid system of "willingness-bid price separation" is adopted to enhance the autonomy and stability of the system; In the willingness decision-making stage of the first level, the agent node calculates the local comprehensive score according to the available bandwidth, current delay, reliability and operating cost parameters, to determine whether to participate in the bid; The scoring function is defined as: wherein, wherein, B i , L i , R i , C i respectively represent the bandwidth, latency, reliability and unit cost of the i-th agent node; α, β, γ, δ are weight parameters configured by the system to balance performance and overhead; when the node score is higher than the set threshold K, the node is determined to participate, and the corresponding willingness flag is set to participate; otherwise, it is set to refuse; for the nodes in the recommended combination, the system will accordingly lower the threshold to encourage them to participate in the combination bidding, thereby improving the feasibility and diversity of the recommended combination; In the bid price generation stage of the second level, only when the node is marked as participating in the willingness layer, the actual bid price calculation process is entered; the generation of bid price considers the local execution cost, expected income and combination synergy factor; when multiple nodes bid in combination, the combination bid price follows the following average concession model: Where n is the number of combination members, and λ is the concession coefficient; this model realizes fair discount within the combination through weighted average of the bid, so that the overall bid maintains balance while reducing cost; after the generation of the willingness and bid, each agent node returns the bid information to the central control node in a standardized response format; the system forms a complete set of bid results at this stage, providing input basis for subsequent bid evaluation and winning determination.

7. The wireless resource allocation method of automatic bidding according to claim 1, characterized in that, The central control node starts the bid evaluation process after collecting all bid responses, and the central control node uniformly organizes the bid information returned by each candidate node or combination, and performs differentiated scoring and determination according to the transfer mode; the system supports two main transfer modes: split mode and redundant mode, which are suitable for high-bandwidth services and high-reliability services, respectively; In the split mode, the task flow is divided into several bandwidth sub-flows and allocated to different agent nodes for parallel transmission; at this time, the system focuses on the total bandwidth and load balancing of the combination, and the combination parameter calculation is as follows: That is, the bandwidth of the combination takes the sum, the delay takes the maximum value, and the reliability takes the minimum value, to ensure that the overall task can be completed when the slowest node is completed; In the redundant mode, multiple nodes transmit the same data simultaneously to improve reliability, and the system focuses on stability and delay instead of bandwidth, and the parameter calculation is as follows: That is, the bandwidth of the combination takes the maximum value, the delay takes the minimum value, and the reliability is obtained by cumulative probability; The system uses a comprehensive score function for unified sorting, and the single node or combination node with the highest score wins the bid; The comprehensive score function is: wherein Q j represents the task satisfaction degree, S j represents the service level agreement score, P j is the bid cost, w1, w2, and w3 are system score weights; the single node or combination with the highest final score is determined as the winning bidder; the central control node generates a task contract and broadcasts the winning result, the winning bidder locks the resources and executes the task, and the non-winning bidder enters a standby or adjustment state; the whole bidding and winning decision-making process is completed within milliseconds, ensuring that the system can still achieve efficient and stable resource allocation in a complex heterogeneous environment.

8. The wireless resource allocation method of automatic bidding according to claim 1, characterized in that, The QoE-QoS mapping and reverse optimization module is introduced in the task modeling stage; this module establishes a mapping model between QoS and QoE through machine learning, automatically deduces the optimal QoS parameter set when the target QoE is specified, and uses the result as a new bid input to realize experience-driven resource scheduling; the system first continuously collects key QoS parameters of each resource agent node during task execution, including bandwidth B, delay L, and reliability R, and generates an experience quality score (standard MOS) E in combination with user feedback or task completion degree; using historical sample data, the system builds a QoS-QoE mapping model to establish the following prediction relationship: E = f(B, L, R) (5) Wherein, f(·) is a trained nonlinear function; according to the model comparison result, the system adopts a LightNN-GBDT hybrid regression model to realize QoE prediction; the model has the nonlinear expression ability of neural network and the stability of gradient boosting tree, and can meet the real-time demand of online reasoning while maintaining the accuracy; in the backward optimization phase, when the system obtains the target experience value E target , the optimal QoS parameter combination (B * , L * , R * ) is solved through a Bayesian optimization algorithm, so that: f(B * ,L * ,R * )≈E target (6) The obtained parameter combination is passed to a combination generation module as a new round of task requirement input; through this mechanism, the system can adaptively adjust the bandwidth allocation, delay constraint and reliability target according to the user experience feedback, so as to realize the dynamic balance between QoS and QoE in different service scenarios; in the task running process, the module also updates the QoE prediction result in real time and calculates the experience deviation; when it is detected that the difference between the current experience value E t and the target value E target exceeds the threshold value ΔE th , the central control node automatically triggers a new round of combination generation and scheduling process, realizing experience-driven closed-loop resource optimization.

9. The method of claim 7, wherein the step of automatically allocating the wireless resources comprises the step of: automatically allocating the wireless resources to the bidders based on the bid prices and the bid quantities. In the initial stage of pre-bid, the application adopts a screening mechanism of "Dual-Pool Layout" to divide the candidate agent nodes into separate bidding pool and combined bidding pool. The system first calculates the comprehensive score value F of each accessible node according to formula 4 i (weight combination {w1, w2, w3} takes the value of 0.4, 0.3, 0.3); the system divides the actual boundary threshold of dual-pool according to the score value F0 of task constraint, that is, for separate bidding, when F i >0.8*F0, the node is selected; for combined bidding, when F i >0.5*F0, the node is selected.

Images