Search

CN-122028060-A - Dynamic deployment system and resource optimization method for large-range wireless network

CN122028060ACN 122028060 ACN122028060 ACN 122028060ACN-122028060-A

Abstract

The invention is applicable to the technical field of wireless networks, and provides a large-range wireless network dynamic deployment system and a resource optimization method. The wireless Mesh backhaul topology is quickly constructed and access parameter initialization is completed under the conditions of limited deployment time and limited backhaul link resources through a staged ad hoc network protocol, future user density levels of coverage areas of all access points are predicted through a user density prediction module, a prediction result, indexes such as channel utilization rate, backhaul load and the like jointly form a multi-agent reinforcement learning state space, and working channels, transmitting power, beam directions and beacon frame transmitting intervals of all access points are jointly optimized through an improved MADDPG model. The invention can improve the total throughput of the whole network and reduce the occurrence rate of congestion in a hot spot area.

Inventors

  • ZHOU YONGHUA

Assignees

  • 深圳通康创智技术有限公司

Dates

Publication Date
20260512
Application Date
20260205

Claims (10)

  1. 1. The large-range wireless network dynamic deployment system is characterized by comprising a central control unit, a plurality of intelligent access points, a user density prediction module, a phased ad hoc network protocol module and a reinforcement learning resource scheduling module, wherein: The phasing ad hoc network protocol module is configured to control each intelligent access point to sequentially execute a discovery and registration stage, a topology and channel planning stage and an access parameter initialization stage after power-up, wherein the topology and channel planning stage adopts an improved minimum spanning tree algorithm based on pre-screening of candidate links of a preset signal quality threshold value and combined with parallel calculation pruning so as to construct a wireless Mesh backhaul topology; the user density prediction module is configured to predict the user density grade and the corresponding prediction confidence coefficient of the future period of the coverage area of each intelligent access point by adopting a combined model of sliding window average, exponential smoothing and long-short-period memory neural network based on the historical user association number time sequence acquired by each intelligent access point, and output a prediction result to the reinforcement learning resource scheduling module; The reinforcement learning resource scheduling module is configured to construct a state space containing real-time network states of all intelligent access points and a user density prediction result by adopting an improved multi-agent depth deterministic strategy gradient algorithm, define an action space containing control actions for adjusting at least one of a working channel, transmitting power, beam direction and beacon frame transmission interval, and jointly optimize wireless parameters of all intelligent access points by means of centralized training and distributed execution based on a reward function containing system total throughput, worst user time delay, configuration switching times, interference suppression indexes and prediction confidence indexes; The central control unit is further configured to trigger updating of the wireless Mesh backhaul topology when a preset condition is met through backhaul load threshold linkage logic based on an optimization result output by the reinforcement learning resource scheduling module so as to realize cross-layer collaborative optimization of access layer parameters and backhaul network topology.
  2. 2. A method for dynamic deployment and resource optimization of a large-scale wireless network, which is applied to the large-scale wireless network dynamic deployment system of claim 1, and is characterized in that the method comprises the following steps: After the intelligent access points are powered on, discovery and registration are executed, and information is reported to a central control unit; the central control unit executes an improved minimum spanning tree algorithm to carry out topology planning and channel allocation based on the reported position information and preset link quality conditions to form an initial wireless Mesh backhaul topology; Based on historical user data collected by each intelligent access point, generating user density grades and prediction confidence of each area in a future period by combining a prediction model; Constructing a state space and an action space of reinforcement learning, wherein the state space at least comprises a real-time channel utilization rate, a user association number, a neighbor channel list, an average packet loss rate, a load of a return father node, a user density level and a prediction confidence coefficient of each intelligent access point, and the action space comprises adjustment of at least one parameter of a working channel, a transmitting power, a beam direction and a beacon frame transmitting interval; operating an improved MADDPG model, selecting actions according to the current state space, generating a wireless parameter configuration result, judging whether to update the Mesh backhaul topology based on backhaul load threshold linkage logic, and issuing the result to each intelligent access point for execution; And continuously optimizing and adjusting parameters of the improved MADDPG model and the prediction model on line based on network performance feedback.
  3. 3. The system of claim 1, wherein the user density prediction module uses a combined model with a sliding window average window duration of 60 seconds, an exponential smoothing coefficient of 0.3, an lstm neural network input layer of 10 time steps, and a hidden layer of 50 neurons.
  4. 4. The method for dynamic deployment and resource optimization of a large-scale wireless network according to claim 2, wherein the improvement points of the improved MADDPG algorithm comprise that the input layer of the Critic network increases the dimension reflecting the load of the backhaul parent node, and that a topology stability penalty term is introduced into the loss function of the Actor network, wherein the penalty term is positively correlated with the switching times of the backhaul topology in the statistical period.
  5. 5. The system of claim 1, wherein the intelligent access point comprises an antenna array supporting beam forming, and wherein the central control unit is further configured to determine an area with a density level reaching or exceeding a third level as a potential hot spot area according to the user density level output by the user density prediction module, and control the intelligent access point to direct a beam main lobe to the area while narrowing the beam width by 20%.
  6. 6. The method of claim 2, wherein the backhaul load threshold linkage logic is configured to select a node with a load rate lower than 45% of a third predetermined threshold from the candidate backhaul nodes as a new backhaul parent node when the predicted user density level of the target intelligent access point coverage area exceeds a first predetermined threshold and the load rate of the current backhaul parent node exceeds a second predetermined threshold by 85%, and complete the topology handover without a service interruption, and the handover process takes no more than 2 seconds.
  7. 7. The method for dynamic deployment and resource optimization of a large-area wireless network according to claim 2, wherein the reward function R adopted by the reinforcement learning resource scheduling module is expressed as R=α T_sys-β L_min-γ S_cfg+δ I+ε C_pred_avg, wherein T_sys is the total throughput of the system, L_min is the average time delay of the worst 5% users, S_cfg is the configuration switching times, I is an interference suppression index, C_pred_avg is the average prediction confidence coefficient of the whole network, and alpha, beta, gamma, delta and epsilon are weight coefficients.
  8. 8. The method for dynamic deployment and resource optimization of a wide area wireless network according to claim 7, wherein when the average prediction confidence c_pred_avg of the whole network is lower than 0.7 and lasts for 3 statistical periods, a policy update is triggered, and the weight coefficient α is adjusted to 1.2, and epsilon is adjusted to 0.02.
  9. 9. The method for dynamic deployment and resource optimization of wide-range wireless network according to claim 2, wherein said improved MADDPG model adopts a centralized training and distributed execution architecture, wherein during the training phase, the central control unit centrally updates the policy network and the value network based on the state information of the whole network, and during the execution phase, each intelligent access point autonomously selects actions based on the local observation state and the policy network parameters acquired from the central control unit and receives the policy parameter update every 2 minutes.
  10. 10. The method for dynamically deploying and optimizing resources in a large-scale wireless network according to claim 2, wherein when it is detected that the user density level of the area continuously increases in two continuous statistical periods, 1 to 2 statistical periods are advanced, the transmission power gear of the area corresponding to the intelligent access point is increased and is switched to a more idle working channel, and meanwhile, the resource allocation of the intelligent access point in the non-hot area is reduced, so that the access congestion occurrence rate of the hot area is reduced.

Description

Dynamic deployment system and resource optimization method for large-range wireless network Technical Field The invention relates to the technical field of wireless networks, in particular to a large-range wireless network dynamic deployment system and a resource optimization method. Background With the increasing activities of large open-air singing, marathon events, urban festival celebrations and the like, the situation of gathering tens of thousands or even hundreds of thousands of people in a limited place in a short time is becoming more common. In such a scenario, the following typical requirements are imposed on the wireless network: 1. The high user density and the high bandwidth demand are that the unit area user density is far higher than that of scenes such as common shops, parks and the like, the number of the on-line terminals can reach tens of thousands of levels at the same time, and the bandwidth demand of users on services such as video live broadcast, social media, short video uploading and the like is extremely high. 2. The deployment time is extremely short, the network usually needs to complete deployment and debugging within hours before the start of the activity, and the network is quickly dismantled after the end of the activity, so that long-period and fine construction and optimization cannot be performed like a macro base station or a park network. 3. The backhaul condition is limited, namely, the movable site is limited by the terrain, the temporary construction structure and the construction, the wired backhaul links cannot be laid, the backhaul can be carried out only by adopting a limited number of wireless Mesh modes, and the capacity and the reliability of the links are both restricted. 4. The traffic load is strongly and dynamically changed, the crowd presents obvious space-time distribution change in the stages of entrance, centralized watching, rest, departure and the like, the network load has strong non-stationary characteristics in time and space, and the traditional optimization strategy designed for a long-term stable scene is difficult to apply. 5. The environmental uncertainty is high, temporary outdoor activities are often influenced by weather, emergencies and the like, so that crowd distribution modes are suddenly changed, and the traditional fixed threshold and static models are difficult to adapt. In the prior art, a plurality of independent research and commercial schemes are respectively available for wireless Mesh topology optimization, AP transmitting power and channel scheduling, beam forming coverage enhancement and wireless resource management based on machine learning, but the schemes have the common defects that 1. The optimization targets are single, most schemes only locally optimize single indexes in throughput, coverage or interference, and the comprehensive trade-off of the whole network capacity, edge experience and configuration stability is lacked. 2. The control dimension is scattered, a unified decision frame is lacking, the topology planning, the access layer parameter adjustment and the beam direction control are designed independently in the existing scheme, and the topology planning, the access layer parameter adjustment and the beam direction control are independent and cannot be optimized cooperatively under the unified state sensing and decision frame. 3. The existing optimization scheme is mostly aimed at macro base stations, park networks or indoor enterprise networks deployed for a long time, and special constraint conditions of short temporary deployment time, limited wireless backhaul, limited equipment quantity and the like are not fully considered. 4. The method lacks a prediction and self-adaptation mechanism for strong dynamic and high uncertainty user behaviors, is mainly used for off-line planning or optimization in the long-term statistics sense even if a part of schemes are introduced into a machine learning method, and lacks a specific prediction and on-line self-adaptation capability for strong transient impacts caused by rapid aggregation and migration of large-scale people in a short time. 5. The model combination is stiff and lacks of self-adaptive weight adjustment, namely the existing prediction model mostly adopts a fixed combination mode, and the weights of all sub-models cannot be dynamically adjusted according to real-time prediction errors, so that the prediction accuracy is greatly reduced when the environment is suddenly changed. Therefore, there is a need to provide a system for dynamic deployment of a wide-range wireless network and a resource optimization method, which aim to solve the above problems. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide a large-range wireless network dynamic deployment system and a resource optimization method so as to solve the problems existing in the background art. The invention is realized in such a way that a large-