A resource management method for a distributed wireless sensor network

Abstract

The application relates to a resource management method for a distributed wireless sensor network. The method comprises the following steps: acquiring channel information of communication between self-organizing network intelligent sensing devices in a power transmission line scene and resource conflict relationships between channels; the resource conflict relationships comprise power interference relationships and frequency spectrum overlap relationships; a conflict hypergraph is constructed based on the channel information and the resource conflict relationships; based on an adjacency matrix of the conflict hypergraph, a signal-to-noise ratio corresponding to each channel in a preset time period, a remaining available resource quantity in the power transmission line scene, and a resource conflict degree between the channels, a resource allocation model trained in advance is used to obtain a resource allocation strategy corresponding to each of the multiple channels, and the resource allocation strategy is used for resource allocation of the multiple channels. The method can solve the interference problem between sensor nodes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless sensor network, and particularly relates to a resource management method for distributed wireless sensor network. BACKGROUND

[0002] In recent years, China has attached great importance to digital development and has started the vigorous promotion of "Digital China" construction, and has put forward higher requirements for the application of digital technology including intelligent sensing in the field of electric power industry. In June 2017, the grand vision of the next generation of smart grid, "transparent grid", was proposed. The future of the power grid is a transparent grid, which effectively integrates sensor technology, information technology, data communication technology, electronic control technology, artificial intelligence, etc. in the power system, and can realize the transparency of device state, operation state and transaction state.

[0003] At present, the main online monitoring system adopts a large number of discrete, single sensors to monitor the operation state of the power system, which cannot realize effective integration and information sharing at the sensor level, so that the development of multi-parameter information fusion sharing and comprehensive diagnosis technology is slow, and it cannot meet the development requirements of smart grid. The self-organizing network intelligent sensing device provides watt-level power output through multiple sensing power supply modules to ensure the real-time online of the self-organizing network communication device, and adopts a technical scheme of miniaturized integration, multi-physical quantity synchronous detection, full-scale measurement and on-demand analog / digital mode to integrate sensitive elements, signal conditioning circuit, microprocessor and communication interface module together, which has the intelligent trend of multi-environment physical quantity sensing and sensor information fusion. The intelligentization and informatization of smart grid need to configure a large number of sensors and data acquisition modules, how to organically integrate sensors and corresponding data acquisition and transmission systems to form a multi-physical quantity integrated sensor array and multi-dimensional information comprehensive analysis technology for efficient state analysis and fault diagnosis of equipment is a hot research direction of the current digital power grid construction.

[0004] Currently, the geographical location and environmental conditions of the transmission line are special, the tower point is multiple, the line is long, and the coverage is wide. It is exposed in the wild all year round. In addition to being attacked by natural disasters such as strong winds, lightning, and heavy rain, it is also often affected by human factors. In recent years, with the rapid development of urban and rural economy, the total mileage of China's transmission lines increased rapidly from 1.15 million kilometers in 2014 to more than 1.59 million kilometers in 2020. Under complex climate and environment, long-distance overhead transmission lines are prone to abnormal failures such as forest fires, tree barriers, and external damage. Tripping events caused by external damage occur frequently, seriously affecting the safe operation of the transmission line. The contradiction between the rapid growth of line mileage in power grid operation and maintenance and the relatively insufficient number of power grid operation and maintenance personnel gradually emerges. In order to effectively prevent the safety hazards of construction and other transmission lines, the transmission line operation and maintenance unit has carried out coverage of key sections of transmission line passages through video / image monitoring terminals since this year, realizing remote online patrol of safety hazards in transmission line passages. At present, video monitoring coverage has been realized for most transmission lines in many regions, and line video coverage has been realized in many regions. Key sections of 500kV and above lines in the network have also been basically fully covered. Through video online patrol, effective prevention of important line construction and other safety hazards is realized, and the efficiency of transmission line operation and maintenance is improved. With the advancement of smart grid construction and the development of artificial intelligence technology, transmission line inspection has transitioned from traditional "human patrol" to "human patrol + machine patrol". However, the flight capability of unmanned aerial vehicles is limited by weather and battery life, and there are problems such as limited monitoring range, high technical requirements for operators, short patrol distance, and influence of no-fly zones, which cannot fully and timely cover the monitoring needs of transmission lines. There are problems such as how to realize data fusion of multiple integrated sensors and remote transmission back to the server.

[0005] Therefore, in order to further improve the observability of power grid operation and enhance the information perception ability of transmission lines, it is necessary to study multi-sensor integration technology to realize all-around and multi-angle monitoring of transmission lines. Through the study of multi-physical quantity integrated sensor technology, the massive deployment of integrated sensing elements is realized. Through the study of multi-physical quantity integrated sensor distributed self-organizing network data communication technology, networking communication between multiple multi-physical quantity integrated sensor devices is completed. Through the study of multi-physical quantity data analysis, feature extraction technology and auxiliary decision-making technology of integrated sensors, the development of multi-physical quantity data fusion application system, and the development of transmission line dynamic capacity monitoring software module, the application of artificial intelligence technology in massive monitoring data of transmission lines is promoted, the actual needs of transmission line operation and maintenance are solved, the measurement data is efficiently and timely transmitted back to the server for processing, and the real-time monitoring of the working state of the transmission line is realized. The goal of automatic, information-based, and digitalized transmission line monitoring is achieved, and the goal of transparent perception of the power grid is ultimately realized.

[0006] In recent years, with the rapid growth of wireless intelligent terminals in smart grids and the increasing demand for mobile data traffic, wireless sensor networks (WSN) have become an integral part of transmission line monitoring systems. In transmission line monitoring, the application of wireless multi-physical integrated sensors is not limited to the measurement of traditional electrical parameters such as current and voltage, but also extends to temperature, humidity, wind speed, vibration, and corrosion levels. Such comprehensive monitoring helps to promptly identify potential fault points, such as insulator aging, conductor slackness, or overheating, and thus take preventive maintenance measures to avoid accidents. In addition, such sensors usually have low power consumption design, suitable for long-term deployment in outdoor environments, reducing the need for frequent battery replacement. Due to the self-organizing nature of distributed nodes, wireless sensor networks are widely used in remote areas with limited network resources. Among them, self-organizing network technology is one of the keys to achieving efficient wireless sensor networks. It allows sensor nodes to automatically establish connections and form networks without a central controller, and can quickly adapt even when the network topology changes, maintaining stable service quality. This capability is crucial for improving the robustness and flexibility of the entire system, especially in the face of natural disasters or other emergencies that cause partial infrastructure damage. However, as the density of sensor nodes increases and communication frequency continues to rise, spectrum resources become increasingly scarce, and interference problems caused by spectrum reuse become increasingly serious. This poses a major challenge to wireless sensor networks to provide low-latency and high-reliability services. In order to solve the interference problem between sensor nodes in distributed wireless sensor networks, an effective resource management method is urgently needed. SUMMARY

[0007] Therefore, it is necessary to provide a resource management method for a distributed wireless sensor network to solve the interference problem between sensor nodes.

[0008] In a first aspect, the present application provides a resource management method for a distributed wireless sensor network, comprising:

[0009] obtaining channel information for communication between self-organizing intelligent sensor devices in a transmission line scenario, and resource conflict relationships between channels; the resource conflict relationships include power interference relationships and spectrum overlap relationships;

[0010] based on the channel information and the resource conflict relationships, constructing a conflict hypergraph;

[0011] The resource allocation model is trained based on the conflict hypergraph adjacency matrix, the signal-to-noise ratio of each channel in a preset time period, the number of remaining available resources in the power line scene, and the resource conflict degree between channels.

[0012] In one of the embodiments, the step of obtaining the resource conflict relationship between the channels comprises:

[0013] For any pair of channels, the first power interference quantity is determined based on the transmission power of the ad hoc network intelligent sensing device on the first channel and the distance between the ad hoc network intelligent sensing devices on the first channel and the second channel; the second power interference quantity is determined based on the transmission power of the ad hoc network intelligent sensing device on the second channel and the distance; if at least one of the first power interference quantity and the second power interference quantity satisfies the power interference condition, it is determined that the first channel and the second channel have a power interference relationship;

[0014] The frequency spectrum overlap quantity is determined based on the center frequency and bandwidth of the first channel and the center frequency and bandwidth of the second channel; if the frequency spectrum overlap quantity satisfies the frequency spectrum overlap condition, it is determined that the first channel and the second channel have a frequency spectrum overlap relationship;

[0015] At least one of the power interference relationship and the frequency spectrum overlap relationship is taken as the resource conflict relationship between the first channel and the second channel.

[0016] In one of the embodiments, the training process of the resource allocation model comprises:

[0017] The sample conflict hypergraph adjacency matrix, the sample signal-to-noise ratio of each channel in a preset time period, the number of sample available resources, and the sample resource conflict degree are obtained, and the initial neural network model is used to obtain the predicted resource allocation strategy corresponding to each of the plurality of channels;

[0018] The model loss is calculated based on the experience resource allocation strategy, the predicted resource allocation strategy, and a preset loss function.

[0019] The model parameters of the initial neural network model are adjusted based on the model loss until a preset stop condition is met, and the trained resource allocation model is obtained.

[0020] In one of the embodiments, the model parameters of the initial neural network model are adjusted based on the model loss until a preset stop condition is met, and the trained resource allocation model is obtained, which comprises:

[0021] The reward value corresponding to the predicted resource allocation strategy is obtained based on a preset reward function.

[0022] The model parameters of the initial neural network model are adjusted based on the reward value and the model loss until a preset stop condition is met, and a trained resource allocation model is obtained.

[0023] In one of the embodiments, the model parameters of the initial neural network model are adjusted based on the model loss until a preset stop condition is met, and a trained resource allocation model is obtained, including:

[0024] The parameter optimization condition is obtained; the parameter optimization condition includes at least one of the maximum resource throughput of each channel, the conflict-free resource allocation, the transmission power of each sensor device being in a preset power range, and the maximum allocation of one resource block for each channel;

[0025] The model parameters of the initial neural network model are adjusted based on the model loss and the parameter optimization condition until a preset stop condition is met, and a trained resource allocation model is obtained.

[0026] In one of the embodiments, the model parameters of the initial neural network model are adjusted based on the model loss until a preset stop condition is met, and a trained resource allocation model is obtained, including:

[0027] The value data corresponding to the predicted resource allocation strategy is obtained based on the preset action value function.

[0028] The model parameters of the initial neural network model are adjusted based on the value data and the model loss until a preset stop condition is met, and a trained resource allocation model is obtained.

[0029] In one of the embodiments, the conflict hypergraph is constructed based on the channel information and the resource conflict relationship, including:

[0030] The channels indicated by the channel information are taken as vertices, and the resource conflict relationship between the channels is taken as edges; the hyperedges are determined based on the channels that have the resource conflict relationship with each other, and the conflict hypergraph is constructed based on the vertices, the edges, and the hyperedges.

[0031] In a second aspect, the application further provides a resource management device for a distributed wireless sensor network, including:

[0032] The acquisition module is configured to acquire channel information for communication between self-organizing network intelligent sensor devices in a power transmission line scenario, and a resource conflict relationship between the channels; the resource conflict relationship includes a power interference relationship and a spectrum overlap relationship.

[0033] The construction module is configured to construct a conflict hypergraph based on the channel information and the resource conflict relationship.

[0034] The allocation module is configured to obtain, based on an adjacency matrix of the conflict hypergraph, a signal-to-noise ratio of each channel in a preset time period, a number of remaining available resources in the power line scene, and a resource conflict degree between the channels, a resource allocation strategy corresponding to each of the plurality of channels by using a pre-trained resource allocation model, and use the resource allocation strategy to allocate resources to the plurality of channels.

[0035] In a third aspect, the present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0036] obtain channel information for communication between the self-organizing network intelligent sensing devices in the power line scene and a resource conflict relationship between the channels; the resource conflict relationship comprises a power interference relationship and a spectrum overlap relationship;

[0037] construct a conflict hypergraph based on the channel information and the resource conflict relationship;

[0038] obtain, based on an adjacency matrix of the conflict hypergraph, a signal-to-noise ratio of each channel in a preset time period, a number of remaining available resources in the power line scene, and a resource conflict degree between the channels, a resource allocation strategy corresponding to each of the plurality of channels by using a pre-trained resource allocation model, and use the resource allocation strategy to allocate resources to the plurality of channels.

[0039] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program implements the following steps when executed by a processor:

[0040] obtain channel information for communication between the self-organizing network intelligent sensing devices in the power line scene and a resource conflict relationship between the channels; the resource conflict relationship comprises a power interference relationship and a spectrum overlap relationship;

[0041] construct a conflict hypergraph based on the channel information and the resource conflict relationship;

[0042] obtain, based on an adjacency matrix of the conflict hypergraph, a signal-to-noise ratio of each channel in a preset time period, a number of remaining available resources in the power line scene, and a resource conflict degree between the channels, a resource allocation strategy corresponding to each of the plurality of channels by using a pre-trained resource allocation model, and use the resource allocation strategy to allocate resources to the plurality of channels.

[0043] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, and the computer program implements the following steps when executed by a processor:

[0044] obtain channel information of communication between self-organizing network intelligent sensing devices in a power transmission line scene, and resource conflict relations between channels; the resource conflict relations include power interference relations and spectrum overlap relations;

[0045] construct a conflict hypergraph based on the channel information and the resource conflict relations;

[0046] obtain resource allocation strategies corresponding to the multiple channels respectively by using a pre-trained resource allocation model based on an adjacency matrix of the conflict hypergraph, signal-to-noise ratios corresponding to the channels respectively in a preset time period, a remaining available resource quantity in the power transmission line scene, and a resource conflict degree between the channels, and the resource allocation strategies are used for resource allocation of the multiple channels.

[0047] The resource management method and device for the distributed wireless sensing network, the computer device, the computer readable storage medium, and the computer program product, by obtaining channel information of communication between self-organizing network intelligent sensing devices in a power transmission line scene, and resource conflict relations between channels, constructing a conflict hypergraph based on the channel information and the resource conflict relations, since the resource conflict relations include power interference relations and spectrum overlap relations, an adjacency matrix of the conflict hypergraph can be used to represent conflict situations between the channels, so as to convert the conflict problem between the channels into a vertex coloring problem of the conflict hypergraph, which is conducive to reducing the computational complexity, obtaining resource allocation strategies corresponding to the multiple channels respectively by using a pre-trained resource allocation model based on an adjacency matrix of the conflict hypergraph, signal-to-noise ratios corresponding to the channels respectively in a preset time period, a remaining available resource quantity in the power transmission line scene, and a resource conflict degree between the channels, and the resource allocation strategies are used for resource allocation of the multiple channels, which is conducive to reducing signal overlap interference in a dense sensor device distribution scene, and improves the reliability of signal transmission. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0049] Figure 1 An application environment diagram of a resource management method for a distributed wireless sensing network in an embodiment;

[0050] Figure 2 A flowchart of a resource management method for a distributed wireless sensing network in an embodiment;

[0051] Figure 3Fig. 1 is a schematic diagram of a WSN network framework in one embodiment;

[0052] Figure 4 Fig. 2 is a schematic diagram of a communication network topology in one embodiment;

[0053] Figure 5 Fig. 3 is a schematic diagram of a conflict hypergraph in one embodiment;

[0054] Figure 6 Fig. 4 is a convergence graph under different proportions of users in one embodiment;

[0055] Figure 7 Fig. 5 is a convergence graph under different convergence rounds in one embodiment;

[0056] Figure 8 Fig. 6 is a schematic diagram of maximum network throughput in one embodiment;

[0057] Figure 9 Fig. 7 is a schematic diagram of average network throughput in one embodiment;

[0058] Figure 10 Fig. 8 is a structural block diagram of a resource management apparatus for a distributed wireless sensor network in one embodiment;

[0059] Figure 11 Fig. 9 is an internal structural diagram of a computer device in one embodiment. DETAILED DESCRIPTION

[0060] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0061] The resource management method for a distributed wireless sensor network provided by the embodiments of the present application can be applied to, for example, Figure 1The application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required by the server 104 to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. The terminal 102 obtains the channel information between the self-organizing network intelligent sensing devices in the power transmission line scene, and the resource conflict relationship between each channel; the resource conflict relationship includes power interference relationship and spectrum overlap relationship; based on the channel information and the resource conflict relationship, a conflict hypergraph is constructed; based on the adjacency matrix of the conflict hypergraph, the signal-to-noise ratio corresponding to each channel in the preset time period, the number of remaining available resources in the power transmission line scene, and the degree of resource conflict between each channel, through the resource allocation model trained in advance, the resource allocation strategy corresponding to each of the multiple channels is obtained, and the resource allocation strategy is used for resource allocation of the multiple channels. Among them, the terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices, the Internet of Things devices can be smart speakers, smart televisions, smart air conditioners, smart car-mounted devices, projection devices, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 can be a standalone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0062] In one exemplary embodiment, as shown in Figure 2 , a resource management method for a distributed wireless sensor network is provided, and the method is applied to the terminal 102 in Figure 1 for example, including the following steps 202 to 206. Among them:

[0063] Step 202, obtaining channel information between self-organizing network intelligent sensing devices in a power transmission line scene, and resource conflict relationship between each channel; the resource conflict relationship includes power interference relationship and spectrum overlap relationship.

[0064] Among them, the self-organizing network intelligent sensing device refers to a sensor device integrated with multiple types of parameter measurement functions in the power transmission line scene. For example, it can integrate the measurement of traditional electrical parameters such as current and voltage, and also extend to temperature, humidity, wind speed, vibration, and corrosion degree. The self-organizing network intelligent sensing device adopts a technical solution of miniaturized integration, multi-physical quantity synchronous detection, full-scale measurement, and on-demand analog / digital mode provision, integrates sensitive elements, signal conditioning circuits, microprocessors, and communication interface modules, and has the intelligent trend of multiple environmental physical quantity sensing and sensor information fusion. Such comprehensive monitoring helps to timely discover potential fault points, such as insulator aging, conductor relaxation or overheating, etc., so as to take preventive maintenance measures to avoid accidents. In addition, such sensors usually have low power consumption design, suitable for long-term deployment in outdoor environment, reducing the need for frequent battery replacement.

[0065] The channel information refers to the information of the channel for communication between the self-organizing network intelligent sensing devices. The channel information can indicate the channel adopted for communication.

[0066] As Figure 3 The figure shows a schematic diagram of a WSN network framework in an embodiment. For example, in the WSN network model in the dense urban scenario, multiple sensor nodes SN establish a self-organizing network in an S2S communication mode, and multiple sensor nodes SN interact with the base station (BS) in an S2I communication mode, so as to send the collected information to the cloud server and accept the resource allocation strategy calculated by the cloud server.

[0067] The channel for communication between sensor nodes is denoted as SL. The interference to the channel SL mainly comes from background noise and signals from SNs allocated with the same channel. The signal conforms to Rayleigh fading, with an average value μ and a path loss factor α. Therefore, the signal-to-interference-plus-noise ratio (SINR) value of the mth SL can be defined as:

[0068]

[0069] where Pm and Pn represent the transmission power of the mth and nth sensor devices, respectively, σ 2 is the constant noise power, h m is the power gain of the channel corresponding to the mth S2S pair, N m is the set of adjacent conflicting users of user m, and h n,m is the interference power gain from the nth sensor device, N S2S denotes the number of S2S pairs. Therefore, the available rate of the mth S2S pair at time t can be calculated as:

[0070]

[0071] where B is a channel bandwidth.

[0072] Based on the directed graph theory, a directed graph is constructed according to the communication relationship between SNs in the WSN network, and the directed graph is represented as an adjacency matrix, as shown in Figure 4 Fig. 1 shows a communication network topology graph in an embodiment. Each node in the communication network topology graph represents an SN, and each edge represents an SL. In the figure, there are 11 SNs and 14 SLs in the communication network topology graph, where the SNs are represented as {S1, S2,..., S11}, and the SLs are represented as {SL1, SL2,..., SL14}. The adjacency matrix of the communication network topology graph can be represented as: 11 14

[0073]

[0074] The communication network topology graph can be represented as a graph where V is a set of nodes, and E is a set of SLs. For each e n ∈ E, there is e i = {(v j ,v i ): v j ,v n ∈ e i}, where the direction of the edge in the ordered pair (v j ,v i ) is determined by the order of the elements, representing a directed edge from v j to v j . For clarity, v T and v i are represented as the first element and the second element of the set e i , respectively. For the communication network topology graph in , the communication relationship between SNs and the transmission direction of SLs can be represented by the adjacency matrix G i .

[0075] Figure 4

[0076]

[0077] where

[0078]

[0079] ​​​​The resource conflict relationship refers to a conflict of resource blocks allocated between channels. Due to limited communication resources densely deployed in the WSN network, close proximity of a large number of S2S pairs can cause interference caused by different power levels between pairs sharing the same RB within a certain range. Therefore, the resource conflict relationship includes a power interference relationship, which is used to represent the channel resource conflict relationship caused by interference caused by device transmit power. In the case of meeting certain power interference conditions, it is determined that there is a power interference relationship between channels. The power interference relationship can be determined by comparing the signal power and the path loss model.

[0080] Meanwhile, in the case of limited spectrum resources, spectrum overlap can also occur between different channels, resulting in conflicts. Therefore, the resource conflict relationship also includes a spectrum overlap relationship, which is used to represent the spectrum overlap between channels. This spectrum overlap interference can exacerbate resource conflicts and adversely affect the overall performance of the network. Therefore, it is crucial to analyze and model the relationships between SLs in the WSN network and allocate different RBs to the S2S links that cause interference. In the case of meeting certain spectrum overlap conditions, it is determined that there is a spectrum overlap relationship between channels.

[0081] Step 204, based on the channel information and the resource conflict relationship, a conflict hypergraph is constructed.

[0082] Among them, the embodiment of the application proposes to construct a conflict hypergraph between SLs based on two types of conflicts, which is used to accurately describe the channel conflict relationship in the WSN network. In order to reduce the complexity of analyzing the conflict hypergraph, the super edge of the conflict hypergraph can be simplified using the theory of maximum clique, thereby reducing the dimension of the association matrix representing the conflict hypergraph. Finally, the resource allocation problem is solved by solving the vertex coloring problem of the conflict hypergraph. In some embodiments, the channels indicated by the channel information can be used as the vertices of the conflict hypergraph, the resource conflict relationship between each channel can be used as the edge of the hypergraph, and the channels with resource conflict relationship between each other can determine the super edge.

[0083] Step 206, based on the adjacency matrix of the conflict hypergraph, the signal-to-noise ratio corresponding to each channel in a predetermined time period, the number of remaining available resources in the power line scenario, and the degree of resource conflict between each channel, a resource allocation model is obtained. The resource allocation strategy is used to allocate resources to multiple channels.

[0084] Among them, the adjacency matrix refers to a matrix composed of the association relationship between the nodes in the conflict hypergraph. As shown in Figure 5 The schematic diagram of the conflict hypergraph in an embodiment is shown in

[0085]

[0086] The signal-to-noise ratio refers to the signal-to-interference-plus-noise ratio (SINR) of each channel SL. The number of remaining available resources refers to the number of all available resource blocks currently remaining in the power line scenario. The resource conflict degree refers to the conflict degree between densely deployed channels in the entire WSN network.

[0087] The resource allocation model refers to a pre-trained neural network model. In some embodiments, the agent based on the DDQN (Double Deep Q-Network) algorithm.

[0088] For example, the terminal takes the adjacency matrix of the conflict hypergraph, the signal-to-noise ratio of each channel corresponding to each channel in a preset time period, the number of remaining available resources in the power line scenario, and the resource conflict degree between channels as inputs of the resource allocation model, and obtains the resource allocation strategy corresponding to each channel output by the resource allocation model.

[0089] The resource allocation strategy is used to allocate resource blocks to multiple channels, thereby avoiding the interference problem between sensor nodes.

[0090] In the above resource management method for a distributed wireless sensor network, the channel information of the self-organizing network intelligent sensing device in the power line scenario is obtained, and the resource conflict relationship between channels is obtained. The conflict hypergraph is constructed based on the channel information and the resource conflict relationship. Since the resource conflict relationship includes the power interference relationship and the spectrum overlap relationship, the adjacency matrix of the conflict hypergraph can be used to represent the conflict between channels, thereby converting the conflict problem between channels into a vertex coloring problem of the conflict hypergraph, which is conducive to reducing the computational complexity. Based on the adjacency matrix of the conflict hypergraph, the signal-to-noise ratio of each channel corresponding to each channel in a preset time period, the number of remaining available resources in the power line scenario, and the resource conflict degree between channels, a pre-trained resource allocation model is used to obtain a resource allocation strategy corresponding to each of the multiple channels. The resource allocation strategy is used to allocate resources to multiple channels, which is conducive to reducing signal overlap interference in a dense sensor device distribution scenario and improving the reliability of signal transmission.

[0091] In an example embodiment, the acquiring of the resource conflict relationship between the channels comprises: for any pair of channels, determining a first power interference amount based on the transmission power of the self-organizing network intelligent sensor device on the first channel and the distance between the self-organizing network intelligent sensor devices on the first channel and the second channel; determining a second power interference amount based on the transmission power of the self-organizing network intelligent sensor device on the second channel and the distance; determining that the first channel and the second channel have a power interference relationship if at least one of the first power interference amount and the second power interference amount satisfies a power interference condition; determining a frequency spectrum overlap amount based on the center frequency and the bandwidth of the first channel and the center frequency and the bandwidth of the second channel; determining that the first channel and the second channel have a frequency spectrum overlap relationship if the frequency spectrum overlap amount satisfies a frequency spectrum overlap condition; and taking at least one of the power interference relationship and the frequency spectrum overlap relationship as the resource conflict relationship between the first channel and the second channel.

[0092] The resource conflict relationship between the channels can be described by taking the resource conflict relationship between any pair of channels as an example. The any pair of channels includes a first channel and a second channel.

[0093] The first power interference amount is used to represent the power interference amount caused by the transmission power of the self-organizing network intelligent sensor device on the first channel. In some embodiments, the first power interference amount can be represented as follows: wherein the transmission power of the self-organizing network intelligent sensor device SN i on the channel SL i is P i , the distance between the self-organizing network intelligent sensor devices SN i and SN j is d ij , and a is a path loss index.

[0094] The second power interference amount is used to represent the power interference amount caused by the transmission power of the self-organizing network intelligent sensor device on the second channel. In some embodiments, the second power interference amount can be represented as follows: wherein the transmission power of the self-organizing network intelligent sensor device SN j on the channel SL j is P j .

[0095] The power interference condition can be a condition satisfied by the power interference amount. If at least one of the first power interference amount and the second power interference amount satisfies the power interference condition, it is determined that the first channel and the second channel have a power interference relationship. In some embodiments, the power interference condition can be represented as follows:

[0096]

[0097] Or,

[0098]

[0099] The above formula indicates that if the power interference level between the channels exceeds the preset SINR threshold, it is determined that there is a power interference relationship between the channels.

[0100] The frequency overlap amount is used to represent the amount of spectral overlap between channels. In some embodiments, the spectral overlap amount can be represented as: max(0, min(f i +B i / 2, f j +B j / 2) - max(f i -B i / 2, f j -B j / 2)), where the center frequency of the channel SL i is f i , the bandwidth is B i , the center frequency of the channel SL j is f j , and the bandwidth is B j .

[0101] In some embodiments, the spectral overlap condition can be represented as:

[0102] max(0, min(f i +B i / 2, f j +B j / 2) - max(f i -B i / 2, f j -B j / 2)) > OL

[0103] where OL represents a spectral overlap amount threshold.

[0104] If the spectral overlap amount between the channels is greater than 0, it indicates that there is an overlapping part between the channels, otherwise, it indicates that there is no overlapping part. The above formula indicates that if the spectral overlap amount between the channels is greater than the spectral overlap amount threshold, it indicates that there is a spectral overlap relationship between the channels, otherwise, there is no spectral overlap relationship.

[0105] If the channels have any one of the power interference relationship and the spectral overlap relationship, it is determined that the channels have a resource conflict relationship, and therefore, at least one of the power interference relationship and the spectral overlap relationship can be used as the resource conflict relationship between the first channel and the second channel.

[0106] In the embodiment, for any pair of channels, the transmission power of the self-organizing network intelligent sensing device and the distance between the self-organizing network intelligent sensing devices on the channel pair are determined to determine whether there is a power interference relationship between the channels, which is beneficial to reflect the influence of the power interference level on the resource conflict between the channels; the center frequency and bandwidth of the channel are determined to determine whether there is a spectrum overlap relationship between the channels, which is beneficial to reflect the influence of the spectrum overlap on the resource conflict between the channels, and the resource conflict relationship includes at least one of the power interference relationship and the spectrum overlap relationship, which is beneficial to accurately reflect the resource conflict between the channels.

[0107] In an exemplary embodiment, the training process of the resource allocation model includes: obtaining a sample adjacency matrix of a sample conflict hypergraph, a sample signal-to-noise ratio corresponding to each channel in a preset time period, a sample available resource quantity, and a sample resource conflict degree, obtaining a predicted resource allocation strategy corresponding to each of the plurality of channels through an initial neural network model; calculating a model loss based on an experience resource allocation strategy, a predicted resource allocation strategy, and a preset loss function; adjusting the model parameters of the initial neural network model based on the model loss until a preset stop condition is met to obtain a trained resource allocation model.

[0108] In a dense WSN network deployment, resource allocation can cause instability in the DDQN training process. This instability adversely affects the speed and accuracy of the SL agent, mainly due to the large number of decisions required for resource management of sensor devices. To alleviate this situation, FedAvg (Federated Averaging) can be integrated with DDQN to form a FDDQN framework, i.e., the initial neural network model can adopt the FDDQN framework. FDDQN combines the advantages of centralized and distributed learning, improving the efficiency and performance of the model. In this framework, a central server coordinates a group of clients, denoted as M. The FDDQN training process iterates through multiple communication rounds to obtain a trained resource allocation model.

[0109] The terminal inputs the sample adjacency matrix, the sample signal-to-noise ratio corresponding to each channel in a preset time period, the sample available resource quantity, and the sample resource conflict degree into the initial neural network model to obtain a predicted resource allocation strategy corresponding to each of the plurality of channels output by the initial neural network model. The experience resource allocation strategy is used as a training label to calculate the model loss with the predicted resource allocation strategy through a preset loss function. The initial neural network adjusts the model parameters based on the model loss until a preset stop condition is met to obtain a trained resource allocation model.

[0110] Exemplarily, the FDDQN training process iterates the following steps over communication rounds from 1 to T:

[0111] 1. Global model propagation: The model training process requires a large amount of data interaction. To minimize communication overhead and improve training efficiency, after initializing the global model, a sample subset S is randomly selected in each round based on multiple rounds of experiments, where the number of selected samples is S M. Then the initialized global model is distributed to these selected subsets.

[0112] 2. Mode: Each selected subset receives the initialized global model and uses the DDQN algorithm based on the MDP model to determine the training parameters for the mth local model. Exemplarily, the goal is to minimize the loss function through gradient descent, which is defined as:

[0113]

[0114] 3. Model aggregation: After t rounds of training, each agent uploads its trained local model to the server. Then the server performs weighted average aggregation, which is represented as:

[0115]

[0116] By repeating the above steps until the FDDQN global model converges.

[0117] The details of this process are summarized as follows in the FDDQN-based conflict-free resource allocation algorithm.

[0118] Requirements: learning rate a, discount factor g, experience replay memory capacity.

[0119] Ensure: main Q network parameters, target Q network parameters.

[0120] 1. Random initialization: The server randomly initializes the online network parameters and sets the parameters w1 = w2. This step is to ensure the consistency of the initial state.

[0121] 2. Select clients: In each iteration, randomly select Rc clients from all clients to participate. Here it is assumed that there are a total of S.M clients.

[0122] 3. Client local update: For each selected client m belongs to [Rc], perform the following operations, calculate the new parameter value

[0123] 4. Local update end flag: After all clients complete the local update, continue to the next iteration.

[0124] 5. Initialize pre-processing model: This step is to prepare for the subsequent learning task.

[0125] 6. Reset state: Reset the state information s0 of the vehicle and the network state for each S2S agent m. This step is to ensure that the state is consistent at the beginning of each training.

[0126] 7. Multi-step training: For each time step t, select an action using the ε-greedy policy and obtain the reward function R and the next state s'. The ε-greedy policy can balance the relationship between exploration and exploitation.

[0127] 8. Store experience: Store the experience of each interaction in the experience replay, denoted as {s, a, r, s'}.

[0128] 9. Sample experience: Randomly sample a small batch of experience data from the experience replay. These experience data are used to calculate the loss function

[0129] 10. Update main network parameters: Update the parameters wmt of the main network according to the calculated loss function.

[0130] 11. Update target network parameters: Periodically update the parameters wt of the target network.

[0131] 12. End flag: When the specified number of training times is reached or other stopping conditions are met, the entire training process ends.

[0132] Repeat the above steps until the Fedavg-DDQN global model converges. To consider the computational complexity of this process, two aspects need to be considered: model training of the DDQN algorithm used by a single S2S agent and parameter weighted average aggregation of the Fedavg server. The complexity of a single DDQN network depends on its complexity TDDQN and the number of local training K lt . Therefore, the complexity of the locally updated DDQN local model can be represented as O((T D3QN ) × K lt ). In addition, the complexity of model aggregation grows linearly with the number of agents P, which is O(P). Therefore, the total complexity of the FedAvg-DDQN algorithm can be represented as:

[0133]

[0134] In this embodiment, the sample adjacency matrix, the sample signal-to-noise ratio, the number of available resources of the sample, and the degree of resource conflict of the sample are taken as inputs of the initial neural network model, and the experience resource allocation strategy is taken as a training label, so as to train the initial neural network to obtain a trained resource allocation model. The resource allocation model can be used for resource allocation of a channel, which is conducive to avoiding the problem of conflict interference between sensor nodes.

[0135] In an example embodiment, the model parameters of the initial neural network model are adjusted based on the model loss until a preset stopping condition is met, to obtain a trained resource allocation model, including: obtaining a reward value corresponding to the predicted resource allocation strategy based on a preset reward function; adjusting the model parameters of the initial neural network model based on the reward value and the model loss until the preset stopping condition is met, to obtain the trained resource allocation model.

[0136] Wherein, the preset reward function is a function for calculating the reward value, and the reward value is used to reward the predicted resource allocation strategy. By maximizing the reward value, the effectiveness of model training can be determined, which can provide non-conflict and efficient resource allocation for dense deployment within the WSN network.

[0137] The preset reward function needs to consider the maximization of throughput and the minimization of constraint conflicts. In some embodiments, the preset reward function is represented as follows:

[0138]

[0139] Wherein, λ1, λ2, λ3∈(0, 1) represent the weight of each reward in the total reward, and k represents the total number of resources, represents the number of RBs used at a certain time, C f represents a conflict factor, which introduces a penalty into the total reward function.

[0140] After each output of the predicted resource allocation strategy, the reward value is calculated, so as to perform model training based on the reward value and the model loss, which is beneficial to guarantee the effectiveness of the model.

[0141] For example, in order to provide a clear, consistent and calculable framework for the model, it is necessary to construct a corresponding Markov decision process (MDP), including state space, action space, reward function and strategy.

[0142] State space: In the resource allocation environment supporting dense deployment within the WSN network, each S2S link is regarded as an agent. At each time period t, the model learns to make the best choice through interaction with the environment and repeated trials. In particular, the agent observes the current environment and selects the best action. Based on this selection, the model receives the corresponding reward feedback, and then the environment changes again for a new round of cycle. Assuming that the nth S2S link observation is as follows:

[0143]

[0144] Wherein, represents the number of remaining available resources in the WSN network, G H represents the adjacency matrix of the conflict hypergraph, SINR of all SLs at time k, denotes the degree of collision in the dense deployment within the WSN network.

[0145] Action space: The action taken by the model to analyze the environment according to the state space at time step t is denoted as:

[0146]

[0147] wherein, denotes the selection of channel nodes, and denotes the selection of resource blocks.

[0148] Reward function: Once an action A(t) is selected in state r t , a reward r t is obtained according to the reward function. It is worth noting that one of the advantages of the reward is that the effectiveness of the model training can be determined by maximizing the reward value. This allows for collision-free and efficient resource allocation for dense deployment within the WSN network.

[0149] In this embodiment, by calculating the reward value through the preset reward function after outputting the predicted resource allocation strategy each time, the model training is carried out based on the reward value and the model loss, which is conducive to ensuring the effectiveness of the model.

[0150] In one exemplary embodiment, the model parameters of the initial neural network model are adjusted based on the model loss until the preset stopping condition is met, and a trained resource allocation model is obtained, including: obtaining a parameter optimization condition; the parameter optimization condition includes at least one of the maximum resource throughput of each channel, collision-free resource allocation, the transmission power of each sensor device being within a preset power range, and at most one resource block being allocated to each channel; the model parameters of the initial neural network model are adjusted based on the model loss and the parameter optimization condition until the preset stopping condition is met, and a trained resource allocation model is obtained.

[0151] wherein, to ensure efficient communication in the dense deployment within the WSN network, the throughput of the SL link needs to be as large as possible, the embodiments of the present application achieve this by allocating resources in the dense deployment within the WSN network to SL as a hypergraph node coloring problem. In some embodiments, the optimization problem is defined as the following four parameter optimization conditions:

[0152]

[0153] wherein, k indicates that the kth resource block is allocated to the mth S2S link, s.t.C f = 0 represents the condition for achieving collision-free resource allocation, indicates that at most 1 resource block can be selected for allocation to SL, i.e. at most one resource block is allocated to each channel, indicates that the resource throughput of each channel is maximized, indicates that the transmission power of each sensor device is within a preset power range.

[0154] The terminal adjusts the model parameters of the initial neural network model according to the model loss and the parameter optimization condition until the preset stopping condition is met, and obtains the trained resource allocation model.

[0155] In the embodiment, the parameter optimization condition is determined, and the model is trained based on the model loss and the parameter optimization condition, which is beneficial to ensure that the throughput of the channel is as large as possible, thereby ensuring efficient communication.

[0156] In one exemplary embodiment, the model parameters of the initial neural network model are adjusted based on the model loss until the preset stopping condition is met, and the trained resource allocation model is obtained, including: obtaining value data corresponding to the predicted resource allocation strategy based on a preset action value function; adjusting the model parameters of the initial neural network model based on the value data and the model loss until the preset stopping condition is met, and obtaining the trained resource allocation model.

[0157] Wherein, after the proposed conflict-free resource allocation problem for dense deployment within the WSN network is converted into MDP, reinforcement learning (RL) can be used to solve the MDP. In this case, the SL is set as an agent. Q-learning is a model-free algorithm that uses the Bellman equation to iteratively update the value function. This is achieved by continuously selecting actions A and updating the action value function Q through the state space S. Finally, the optimal action value function Q is learned, thereby implementing the optimal strategy. In the dense deployment within the WSN network, the SL resource allocation is performed according to the allocation strategy π, which can be expressed as a preset action value function Q(s, a) according to the Bellman equation, which is defined as:

[0158]

[0159] Wherein, γ t Indicates the discount factor, and the initial state and action are represented by S0 and A0, respectively.

[0160] The terminal can calculate the value data corresponding to the predicted resource allocation strategy based on the preset action value function, and then perform model training based on the value data and the model loss.

[0161] In some embodiments, the terminal can also constantly select the optimal action according to the optimal strategy, and use π * Indicates that the optimal action value function Q *, which represents the optimal long-term return. It is expressed as follows:

[0162]

[0163] Since determining the optimal action value function requires evaluating many actions for each state recorded in the Q-table, the Bellman equation can be used to iteratively compute to simplify this process. In some embodiments, the optimal action value function Q * is defined by the Bellman equation as follows:

[0164]

[0165] In dense deployment scenarios within WSN networks, model-free algorithms (such as Q-learning) are typically employed, which makes it difficult to determine transition probabilities. To address this issue, the temporal difference (TD) method is used to solve the Q * value estimation problem. In some embodiments, the update process at each time step is defined as follows:

[0166] Q(S t ,A t )←Q(S t ,A t )+α[R t+1 +Q(S t+1 ,A t+1 )-Q(S t ,A t )]

[0167] where a is the learning rate. However, in dense WSN network deployments, the large state-action space makes it difficult to store the Q-table, and a deep neural network (DNN) is introduced to iteratively update the weights w m to approximate the Q * table values. In some embodiments, this approximation in deep Q Networks (DQN) is mathematically expressed as:

[0168] y t =r t +γq(s t+1 ,argmax a q(s t+1 ,a;w m );w m )

[0169] In DQN, the action value function Q updates its estimates with the current estimated action values. Due to the correlation of the above phenomenon, the dense deployment of agents within WSN networks makes them prone to overestimation bootstrap during the learning process. This leads to instability and low learning efficiency issues. To avoid bootstrap and implement the DDQN algorithm, a target network can be incorporated, which in some embodiments is defined by the algorithm as follows:

[0170] y t = r t + γq(s t+1 , argmax a q(s t+1 , a; w m ); w t )

[0171] where w t denotes the training network parameters, both the DDQN training network and the target network exist, the former selects the action, and the latter estimates the selected action based on the current state.

[0172] In order to further weaken the correlation of Agent training in the process of intensive deployment of resource allocation in WSN network, and achieve better training effect, the experience replay mechanism can be introduced. The SL agent reads the current state, and selects the corresponding optimal action a * according to the optimal Q value Q * . In addition, in order to avoid falling into a local optimal solution, each SL intelligence adopts an epsilon greedy mechanism to select the optimal action a * with a probability of 1-epsilon and obtain the corresponding reward R. Then, this enables the transition to the next state st+1, completes a round, and stores the experience E=(s, a, r, s t+1 ) in the experience replay memory, marked as This process is called experience replay [X]. Once the experience replay memory is full, a small batch of samples with a size of is selected for training. Then the loss function is calculated by continuously adjusting the estimate of the Q value function through the TD method, the purpose of which is to approximate the optimal Q value function, which can be defined as in some embodiments:

[0173]

[0174] The continuous evaluation of the loss function with respect to the value function allows the model to adapt more effectively to the environment, thereby increasing the probability of achieving higher returns.

[0175] In this embodiment, the value data corresponding to the predicted resource allocation strategy is obtained through the preset action value function, and the model training is performed based on the value data and the model loss, so that the model learns the optimal action value function, thereby facilitating the implementation of the optimal resource allocation strategy.

[0176] In an exemplary embodiment, based on the channel information and the resource conflict relationship, a conflict hypergraph is constructed, including: taking the channels indicated by the channel information as vertices, and taking the resource conflict relationship between the channels as edges; determining a hyperedge based on channels that have a resource conflict relationship with each other, and constructing a conflict hypergraph based on the vertices, edges and hyperedges.

[0177] To manage resource conflicts in WSN networks, it is necessary to model SLs for further analysis. Since the conflict relationships between SLs are complex many-to-many relationships, a hypergraph is introduced as a modeling tool, constructing a conflict hypergraph with SLs as vertices. refer to Figure 5 The conflict hypergraph model in the text illustrates two types of conflict relationships between SLs, where... yes The set of SL in the , which is represented as It is the set of edges representing the conflict relationships of SL.

[0178] This conflict hypergraph can be represented by the incidence matrix H.

[0179]

[0180] Where h ij Indicate vertex v i Does it belong to the superedge e? j :

[0181]

[0182] The incidence matrix H accurately describes the relationship between vertices and hyperedges in a conflict hypergraph.

[0183] Based on the two types of resource conflict relationships between SLs introduced in step 202, quantification can be achieved. Figure 4 The communication topology diagram illustrates the conflict relationships between nodes (SLs). Using SLs as vertices, conflicting vertices are connected by edges, and hyperedges are determined based on channels with pairwise resource conflicts, thus constructing a conflict hypergraph. This method not only accurately represents the many-to-many conflict relationships between SLs but also provides a graph-theory-based mathematical solution for subsequent resource allocation problems.

[0184] refer to Figure 5 The conflict hypergraph is constructed in the figure, in which some SLs are omitted to clearly show the topological relationships between SLs in the conflict hypergraph for subsequent analysis.

[0185] Since all SLs within the same hyperedge are fully connected, meaning their conflict relations are identical, the conflict avoidance problem can be solved by coloring each hyperedge individually. This means the number of hyperedges to be computed significantly impacts computational complexity. This transformation preserves all conflict information while reducing the dimensionality of the hypergraph's incidence matrix, thus avoiding unnecessary computational overhead in handling redundant information. Finally, using hypergraph theory as a mathematical model, the spectrum resource allocation problem is transformed into a more manageable hypergraph vertex coloring problem.

[0186] In this embodiment, the channels indicated by the channel information are taken as vertices, and the resource conflict relationship between the channels is taken as edges; the hyper-edges are determined based on the channels having a resource conflict relationship with each other, and the conflict hypergraph is constructed based on the vertices, edges and hyper-edges. This method not only accurately represents the many-to-many conflict relationship between the channels, but also provides a mathematical solution based on graph theory for the subsequent resource allocation problem.

[0187] To illustrate the resource management method for a distributed wireless sensor network in this scheme in detail and its effects, a most detailed embodiment is described as follows:

[0188] For the resource management scene of the self-organizing network intelligent sensing device in the power line scene, the terminal acquires channel information for communication between self-organizing network intelligent sensing devices in the power line scene, and resource conflict relationships between channels; the resource conflict relationships include power interference relationships and spectrum overlap relationships; a conflict hypergraph is constructed based on the channel information and the resource conflict relationships; based on the adjacency matrix of the conflict hypergraph, the signal-to-noise ratio of each channel corresponding to each channel in a preset time period, the number of remaining available resources in the power line scene, and the resource conflict degree between the channels, a resource allocation model trained in advance is used to obtain a resource allocation strategy corresponding to each of the multiple channels, and the resource allocation strategy is used for resource allocation of the multiple channels.

[0189] Among them, the effectiveness of the proposed algorithm is verified by simulation experiments, numerical experiments are carried out through the example of classification problems of UCI database, and the results of numerical experiments show the effectiveness of the proposed algorithm. The simulation results verify the performance of the proposed strategy; the FedAvg-DDQN algorithm is compared with the greedy-based network resource allocation algorithm, the randomized network resource allocation algorithm and the resource allocation algorithm based on the maximum node degree (MND). The model parameters of the proposed FedAvg-DDQN algorithm and the simulation parameters of the WSN network scene are shown in Table 1.

[0190] Table 1 Simulation parameter table

[0191]

[0192] As Figure 6Fig. 6 is a convergence graph for different discount rates (γ) in an embodiment. As can be seen in the graph, the cumulative reward varies with the number of training rounds for different discount rates (γ). It can be seen that the strategies for all discount rates achieve convergence, with the discount rate of 0.9 performing best, achieving convergence around 150 rounds and having the highest final cumulative reward, close to 360. In comparison, the curves for discount rates of 0.95 and 0.85, while performing relatively smoothly, have slightly lower final cumulative rewards than the former, while the model with a discount rate of 0.99 has the slowest convergence speed, with a final cumulative reward of only around 340. This shows that a discount rate of 0.9 can achieve the best balance in the current scenario in the short term, ensuring both convergence speed and a high cumulative reward.

[0193] As Figure 7 Fig. 7 is a convergence graph for different learning rates (Lr) in an embodiment. As can be seen in the graph, the cumulative reward varies with the number of training rounds for different learning rates (Lr). It can be seen that the strategies for all learning rates achieve convergence, but the choice of learning rate has a significant impact on the convergence speed and cumulative reward of the model. Among them, the learning rate of 0.0003 performs best, with a faster convergence speed, reaching a stable state around 300 rounds, and a final cumulative reward close to 350. The learning rate of 0.0005 also performs well, with a cumulative reward slightly lower than 0.0003. The learning rate of 0.0007 has a faster convergence speed but poor stability, with a slightly lower final cumulative reward. The learning rate of 0.0001 has the slowest convergence speed and the lowest cumulative reward, about 300. This shows that 0.0003 is the best learning rate in the current scenario.

[0194] As Figure 8 Fig. 8 is a schematic diagram of the maximum network throughput in an embodiment. The graph shows the maximum network throughput for different numbers of SLs in a dense WSN network scenario using different algorithms. As the number of SLs increases, the number of available communication channels in the system also increases, resulting in an overall upward trend in the maximum network throughput for the four algorithms. In a dense WSN network scenario with 20 SLs, FedavgDDQN achieves a maximum network throughput of over 80 Mb / s, while all other comparative algorithms are below 70 Mb / s. With 50 SLs, Fedavg-DDQN achieves a maximum network throughput of 180 Mb / s, while the MDQN algorithm ranges from 100 Mb / s to 140 Mb / s. In summary, according to Figure 8 , the FedAvg-DDQN algorithm is superior to the MDQN, FDQN, and SDQN algorithms. The FedAvg-DDQN algorithm provides a significant advantage in enhancing network capacity in dense WSN network scenarios.

[0195] As Figure 9Fig. 6 is a diagram illustrating the average network throughput for one embodiment. The diagram shows the average network throughput for different numbers of SLs in a dense WSN network scenario using different algorithms. As the number of S2S agents increases, the network throughput for all four algorithms shows a significant upward trend. When there are 20 SLs in a dense WSN network scenario, the DDQN algorithm can achieve an average throughput of nearly 80 Mb / s, while the FDQN and SDQN algorithms remain relatively constant. The MDQN algorithm achieves an average throughput of approximately 60 Mb / s. When there are 40 SLs in a dense WSN network scenario, the Fedavg-DDQN algorithm achieves an average throughput of nearly 160 Mb / s, while the other three algorithms have an average throughput of less than 100 Mb / s. In summary, Figure 9 Fig. 6 shows that the FedAvg-DDQN algorithm outperforms the other three algorithms and has a significant advantage in terms of average network throughput in a dense WSN network scenario.

[0196] The resource management method for a distributed wireless sensor network described above, by acquiring channel information for communication between self-organizing network intelligent sensing devices in a power transmission line scenario and resource conflict relationships between channels, constructing a conflict hypergraph based on the channel information and the resource conflict relationships, since the resource conflict relationships include power interference relationships and spectral overlap relationships, the adjacency matrix of the conflict hypergraph can be used to represent the conflict between channels, thereby converting the conflict problem between channels into a vertex coloring problem of the conflict hypergraph, which is conducive to reducing the computational complexity; based on the adjacency matrix of the conflict hypergraph, the signal-to-noise ratio of each channel corresponding to each channel in a preset time period, the number of remaining available resources in the power transmission line scenario, and the degree of resource conflict between channels, a resource allocation model trained in advance is used to obtain a resource allocation strategy corresponding to each of the multiple channels, the resource allocation strategy is used for resource allocation of the multiple channels, which is conducive to reducing signal overlap interference in a dense sensor device distribution scenario and improving the reliability of signal transmission. In addition, the conflict hypergraph is used to model the communication channels in a densely distributed wireless sensor network scenario, the characteristics of the hypergraph are used to integrate different types of conflict channels into a hypergraph, reducing redundant information, solving the vertex coloring problem of the hypergraph to solve the resource allocation problem, and reducing the computational complexity. The clique theory is used to simplify the hyperedges of the hypergraph, and the maximum clique is solved to reduce the redundancy of the conflict information in the hyperedge, which reduces the dimension of the adjacency matrix representing the hypergraph and improves the computational efficiency of resource allocation. In order to solve the node coloring problem in the conflict hypergraph, the problem is converted into a Markov decision process (MDP) model, and the introduced MDP model is solved by the DDQN algorithm, which reduces the amount of calculation and speeds up the training speed. For the F-DDQN algorithm for synchronous update, the simulation results show that the algorithm has a faster convergence speed and higher network throughput.

[0197] It should be understood that although the steps in the flowcharts involved in the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps.

[0198] Based on the same inventive concept, the embodiments of the present application also provide a resource management apparatus for a distributed wireless sensor network for implementing the above-mentioned resource management method for a distributed wireless sensor network. The implementation scheme for solving the problem provided by the apparatus is similar to the implementation scheme described in the above method, so the specific limitations in one or more resource management apparatus embodiments for a distributed wireless sensor network provided below can refer to the limitations of the resource management method for a distributed wireless sensor network described above, which will not be repeated here.

[0199] In one exemplary embodiment, as shown in Figure 10 A resource management apparatus 100 for a distributed wireless sensor network is provided, comprising an acquisition module 120, a construction module 140 and an allocation module 160, wherein:

[0200] The acquisition module 120 is configured to acquire channel information for communication between intelligent sensor devices in an ad hoc network in a power transmission line scenario, and resource conflict relationships between channels; the resource conflict relationships include power interference relationships and spectral overlap relationships;

[0201] The construction module 140 is configured to construct a conflict hypergraph based on the channel information and the resource conflict relationships;

[0202] The allocation module 160 is configured to obtain a resource allocation strategy corresponding to each of the plurality of channels by using a pre-trained resource allocation model based on an adjacency matrix of the conflict hypergraph, a signal-to-noise ratio corresponding to each of the channels in a preset time period, a number of remaining available resources in the power transmission line scenario, and a resource conflict degree between the channels, and the resource allocation strategy is used for resource allocation of the plurality of channels.

[0203] The resource management device for the distributed wireless sensor network, by acquiring channel information of communication between self-organizing network intelligent sensing devices under the power line scene and resource conflict relationship between channels, constructing a conflict hypergraph based on the channel information and the resource conflict relationship, since the resource conflict relationship includes power interference relationship and spectrum overlap relationship, therefore, the adjacency matrix of the conflict hypergraph can be used to represent the conflict between channels, thereby converting the conflict problem between channels into a vertex coloring problem of the conflict hypergraph, which is conducive to reducing the computational complexity; based on the adjacency matrix of the conflict hypergraph, the signal-to-noise ratio corresponding to each channel in the preset time period, the number of remaining available resources under the power line scene, and the resource conflict degree between channels, through the pre-trained resource allocation model, the resource allocation strategy corresponding to each of the multiple channels is obtained, and the resource allocation strategy is used for resource allocation of the multiple channels, which is conducive to reducing the signal overlap interference in the dense sensor device distribution scene, and improving the reliability of signal transmission.

[0204] In one embodiment, during the training process of the resource allocation model, the allocation module 160 is further configured to: acquire a sample adjacency matrix of a sample conflict hypergraph, a sample signal-to-noise ratio corresponding to each channel in a preset time period, a sample available resource quantity, and a sample resource conflict degree, obtain a predicted resource allocation strategy corresponding to each of the multiple channels through an initial neural network model; calculate a model loss based on the experience resource allocation strategy, the predicted resource allocation strategy, and a preset loss function; adjust the model parameters of the initial neural network model based on the model loss until a preset stop condition is met, and obtain the trained resource allocation model.

[0205] In one embodiment, during the training process of the resource allocation model, the allocation module 160 is further configured to: obtain a reward value corresponding to the predicted resource allocation strategy based on a preset reward function; adjust the model parameters of the initial neural network model based on the reward value and the model loss until the preset stop condition is met, and obtain the trained resource allocation model.

[0206] In one embodiment, during the training process of the resource allocation model, the allocation module 160 is further configured to: obtain a reward value corresponding to the predicted resource allocation strategy based on a preset reward function; adjust the model parameters of the initial neural network model based on the reward value and the model loss until the preset stop condition is met, and obtain the trained resource allocation model.

[0207] In an embodiment, the model parameters of the initial neural network model are adjusted based on the model loss until a preset stop condition is met, to obtain the trained resource allocation model, and the allocation module 160 is further configured to: obtain value data corresponding to the predicted resource allocation strategy based on a preset action value function; and adjust the model parameters of the initial neural network model based on the value data and the model loss until the preset stop condition is met, to obtain the trained resource allocation model.

[0208] In an embodiment, based on the channel information and the resource conflict relationship, a conflict hypergraph is constructed, and the construction module 140 is further configured to: take the channels indicated by the channel information as vertices and take the resource conflict relationship between the channels as edges; determine a hyperedge based on channels that have a resource conflict relationship with each other, and construct the conflict hypergraph based on the vertices, the edges, and the hyperedge.

[0209] The various modules in the resource management apparatus for a distributed wireless sensor network described above can be implemented in whole or in part by software, hardware, and combinations thereof. The various modules described above can be embedded in a processor in a hardware form or independent of the processor in a computer device, or can be stored in a memory in a software form in the computer device, so as to be called and executed by the processor to perform the operations corresponding to the various modules.

[0210] In an exemplary embodiment, a computer device is provided, which can be a terminal, and an internal structure diagram of the computer device can be as shown in Figure 11As shown in the figure. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be realized through WIFI, mobile cellular network, near field communication (Near Field Communication, NFC) or other technologies. The computer program is executed by the processor to realize a resource management method for a distributed wireless sensor network. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0211] Those skilled in the art can understand that, Figure 11 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0212] In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in each of the method embodiments described above.

[0213] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps in each of the method embodiments described above.

[0214] In one embodiment, a computer program product is provided, including a computer program, and the computer program is executed by a processor to implement the steps in each of the method embodiments described above.

[0215] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0216] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of each method. In the embodiments provided in the present application, any reference to memory, database or other medium can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (Resistive Random Access Memory, ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (Artificial Intelligence, AI) processor, etc., without being limited thereto.

[0217] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, any combination of these technical features is deemed to be within the scope of the present application.

[0218] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A resource management method for a distributed wireless sensor network, characterized in that, The method comprises: obtaining channel information of communication between self-organizing network intelligent sensing devices in a power transmission line scene, and resource conflict relationships between channels; the resource conflict relationships comprise power interference relationships and spectrum overlap relationships; based on the channel information and the resource conflict relationships, a conflict hypergraph is constructed; based on an adjacency matrix of the conflict hypergraph, a signal-to-noise ratio corresponding to each channel in a preset time period, a number of remaining available resources in the power transmission line scene, and a resource conflict degree between channels, a resource allocation model pre-trained is used to obtain a resource allocation strategy corresponding to each channel, and the resource allocation strategy is used for resource allocation of the channels. The step of obtaining the resource conflict relationships between the channels comprises: for any pair of channels, a first power interference quantity is determined based on the transmission power of the self-organizing network intelligent sensing device on the first channel and the distance between the self-organizing network intelligent sensing devices on the first channel and the second channel; a second power interference quantity is determined based on the transmission power of the self-organizing network intelligent sensing device on the second channel and the distance; if at least one of the first power interference quantity and the second power interference quantity satisfies a power interference condition, it is determined that the first channel and the second channel have a power interference relationship; a spectrum overlap quantity is determined based on the center frequency and bandwidth of the first channel and the center frequency and bandwidth of the second channel; if the frequency overlap quantity satisfies a spectrum overlap condition, it is determined that the first channel and the second channel have a spectrum overlap relationship; at least one of the power interference relationship and the spectrum overlap relationship is used as the resource conflict relationship between the first channel and the second channel.

2. The method of claim 1, wherein, The training process of the resource allocation model comprises: a sample adjacency matrix of a sample conflict hypergraph, a sample signal-to-noise ratio corresponding to each channel in a preset time period, a sample available resource quantity, and a sample resource conflict degree are obtained, and a predicted resource allocation strategy corresponding to each channel is obtained through an initial neural network model; a model loss is calculated based on an experience resource allocation strategy, a predicted resource allocation strategy, and a preset loss function; the model parameters of the initial neural network model are adjusted based on the model loss until a preset stop condition is met, and a trained resource allocation model is obtained.

3. The method of claim 2, wherein, The training process of the resource allocation model comprises: a reward value corresponding to the predicted resource allocation strategy is obtained based on a preset reward function; the model parameters of the initial neural network model are adjusted based on the reward value and the model loss until a preset stop condition is met, and a trained resource allocation model is obtained.

4. The method of claim 2, wherein, The training process of the resource allocation model comprises: obtaining parameter optimization conditions; the parameter optimization conditions include at least one of maximum resource throughput of each channel, collision-free resource allocation, transmission power of each sensor device being in a preset power range, and maximum allocation of one resource block per channel; adjusting model parameters of the initial neural network model based on the model loss and the parameter optimization conditions until a preset stopping condition is met, to obtain a trained resource allocation model.

5. The method of claim 2, wherein, The adjusting of the model parameters of the initial neural network model based on the model loss until the preset stopping condition is met to obtain the trained resource allocation model includes: obtaining value data corresponding to a predicted resource allocation strategy based on a preset action value function; adjusting the model parameters of the initial neural network model based on the value data and the model loss until the preset stopping condition is met, to obtain the trained resource allocation model.

6. The method of claim 1, wherein, The constructing of the conflict hypergraph based on the channel information and the resource conflict relationship includes: taking channels indicated by the channel information as vertices and taking resource conflict relationships between the channels as edges; determining hyper-edges based on channels that have resource conflict relationships with each other; and constructing the conflict hypergraph based on the vertices, edges, and hyper-edges.

7. A resource management apparatus for a distributed wireless sensor network, characterized by, The apparatus includes: an obtaining module, configured to obtain channel information for communication between self-organizing network intelligent sensor devices in a power transmission line scenario and resource conflict relationships between channels; the resource conflict relationships include power interference relationships and spectrum overlap relationships; a constructing module, configured to construct a conflict hypergraph based on the channel information and the resource conflict relationships; an allocating module, configured to obtain resource allocation strategies corresponding to the channels respectively by using a pre-trained resource allocation model, based on an adjacency matrix of the conflict hypergraph, signal-to-noise ratios of the channels respectively in a preset time period, a remaining available resource quantity in the power transmission line scenario, and resource conflict degrees between the channels; the resource allocation strategies are used for resource allocation of the channels. The obtaining module is further configured to, for any pair of channels, determine a first power interference quantity based on a transmission power of a self-organizing network intelligent sensor device on a first channel on the first channel and a distance between the self-organizing network intelligent sensor devices on the first channel and a second channel; determine a second power interference quantity based on a transmission power of a self-organizing network intelligent sensor device on a second channel on the second channel and the distance; determine that the first channel and the second channel have a power interference relationship if at least one of the first power interference quantity and the second power interference quantity satisfies a power interference condition; determine a frequency overlap quantity based on a center frequency and a bandwidth of the first channel and a center frequency and a bandwidth of the second channel; determine that the first channel and the second channel have a spectrum overlap relationship if the frequency overlap quantity satisfies a spectrum overlap condition; and take at least one of the power interference relationship and the spectrum overlap relationship as a resource conflict relationship between the first channel and the second channel.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images