Wireless network optimization method, electronic device, and computer program product

Abstract

The present disclosure provides a wireless network optimization method, an electronic device and a computer program product, and relates to the technical field of computers. The wireless network optimization method of the present disclosure realizes wireless network optimization based on a distributed spiking neural network, and comprises the following steps: acquiring state data of each sensor node in a wireless network, each sensor node corresponding to a neuron node of the spiking neural network; performing feature extraction on the state data of the sensor node to determine a state feature vector of the sensor node; constructing an utility function with the optimization objectives of maximizing the throughput of the wireless network and minimizing the energy consumption of the sensor node based on the state feature vector of the sensor node; determining an optimization strategy of the sensor node by maximizing the utility function; and optimizing the wireless network based on the optimization strategy of the sensor node.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a wireless network optimization method, electronic device, and computer program product. Background Technology

[0002] Current optimization methods for multi-hop wireless networks mostly rely on centralized control or static rule configuration, which not only suffers from single-point failure risks and poor scalability, but also struggles to adapt to complex environments such as dynamic changes in network topology, time-varying channel conditions, and limited node resources. While existing distributed optimization techniques incorporate machine learning models for policy decision-making, they are often based on continuous neural networks, resulting in high computational overhead and power consumption, making them unsuitable for low-power edge nodes, and lacking the ability to effectively model the temporal evolution characteristics of states. Summary of the Invention

[0003] This disclosure provides a wireless network optimization method, an electronic device, and a computer program product.

[0004] According to one aspect of this disclosure, a wireless network optimization method is provided, which optimizes the wireless network based on a distributed spiking neural network. The wireless network optimization method includes: acquiring state data of each sensor node in the wireless network, where each sensor node corresponds to a neuron node of the distributed spiking neural network, and the state data is used to describe the communication capability, energy reserve, and location information of the sensor node in the wireless network; extracting features from the state data of the sensor node to determine the state feature vector of the sensor node; constructing a utility function based on the state feature vector of the sensor node, with the optimization objectives of maximizing the wireless network throughput and minimizing the energy consumption of the sensor node; determining an optimization strategy for the sensor node by maximizing the utility function, the optimization strategy being used to adjust the resource configuration of the sensor node; and optimizing the wireless network based on the optimization strategy of the sensor node.

[0005] Based on one aspect of wireless network optimization, state data of each sensor node in the wireless network is collected. Based on this state data, a state feature vector is determined to describe the communication capability, energy reserves, and location information of each sensor node. This state feature vector is used to construct a utility function that aims to maximize network throughput and minimize node energy consumption. The optimization strategy for each sensor node is then solved by maximizing this utility function. This optimization strategy is used to adjust the resource allocation of the sensor node. Finally, based on the optimization strategies of each sensor node, the overall wireless network is optimized, achieving a synergistic improvement in wireless network performance and energy efficiency, while reducing communication overhead and decision latency.

[0006] A wireless network optimization method according to at least one embodiment of the present disclosure optimizes the wireless network based on the optimization strategy of the sensor node, comprising: optimizing the wireless network based on the optimization strategy of the sensor node, determining new state data of the sensor node; and iteratively optimizing the wireless network based on the new state data of the sensor node until, in a consecutive target number of iterations, the difference between the optimization strategy of the sensor node and the optimization strategy of the sensor node in the previous iteration is less than a target threshold.

[0007] A wireless network optimization method according to at least one embodiment of the present disclosure further includes: after each iteration of the wireless network, calculating the gradient of the utility function of the sensor node relative to the optimization strategy of the sensor node, the gradient representing the changing trend of throughput gain and energy consumption cost under the optimization strategy; and updating the synaptic weights of the distributed spiking neural network through backpropagation based on the gradient.

[0008] According to at least one embodiment of the wireless network optimization method of this disclosure, feature extraction is performed on the state data of the sensor node to determine the state feature vector of the sensor node, including: pulse coding of the state data of the sensor node to determine the pulse sequence of the sensor node; and feature extraction is performed on the pulse sequence of the sensor node to determine the state feature vector of the sensor node.

[0009] According to at least one embodiment of the wireless network optimization method of this disclosure, pulse coding is performed on the state data of the sensor node to determine the pulse sequence of the sensor node, including: obtaining the optimization priorities corresponding to the throughput and energy consumption of the sensor node respectively; and pulse coding is performed on the state data of the sensor node based on the mapping relationship between the optimization priority and the pulse parameters to determine the pulse sequence of the sensor node, wherein the pulse parameters are used to determine the pulse firing frequency and pulse relative phase of the pulse coding.

[0010] According to at least one embodiment of the wireless network optimization method of this disclosure, the optimization strategy of each sensor node in the wireless network is determined based on Nash equalization.

[0011] A wireless network optimization method according to at least one embodiment of the present disclosure optimizes the wireless network based on the optimization strategy of the sensor node, including: decoding the optimization strategy based on the target pulse emission frequency and the target time interval to generate the transmit power, routing selection, and sleep-wake strategy of the sensor node, wherein the routing selection is used to determine the path for the sensor node to transmit to the next sensor node in the wireless network, and the sleep-wake strategy is used to determine the working state of the sensor node; and optimizing the wireless network based on the transmit power, routing selection, and sleep-wake strategy of the sensor node.

[0012] According to at least one embodiment of the wireless network optimization method of this disclosure, a utility function is constructed based on the state feature vector of the sensor node, with the optimization objectives of maximizing wireless network throughput and minimizing sensor node energy consumption. The utility function is constructed by combining the state feature vectors of other sensor nodes in the wireless network with the state feature vectors of the sensor node.

[0013] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform a wireless network optimization method according to any embodiment of this disclosure.

[0014] According to another aspect of this disclosure, a readable storage medium is provided, wherein executable instructions are stored therein, which, when executed by a processor, are used to implement a wireless network optimization method according to any embodiment of this disclosure.

[0015] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a wireless network optimization method according to any embodiment of this disclosure. Attached Figure Description

[0016] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0017] Figure 1 This is a schematic diagram of the overall process of a wireless network optimization method according to one embodiment of the present disclosure.

[0018] Figure 2 This is a flowchart illustrating the process of determining the state feature vector of a sensor node in a wireless network optimization method according to one embodiment of the present disclosure.

[0019] Figure 3 This is a flowchart illustrating the process of determining the pulse sequence of a sensor node in a wireless network optimization method according to one embodiment of the present disclosure.

[0020] Figure 4 This is a schematic flowchart illustrating the optimization of a wireless network according to one embodiment of the present disclosure.

[0021] Figure 5 This is a schematic diagram of the process of iteratively optimizing a wireless network in a wireless network optimization method according to one embodiment of the present disclosure.

[0022] Figure 6 This is a schematic diagram of the backpropagation update process in a wireless network optimization method according to one embodiment of the present disclosure.

[0023] Figure 7 This is a schematic diagram of a wireless network optimization method according to one embodiment of the present disclosure.

[0024] Figure 8 This is a flowchart illustrating a wireless network optimization method according to one embodiment of the present disclosure.

[0025] Figure 9 This is a comparative schematic diagram of the Pareto optimal solution set of a wireless network optimization method according to one embodiment of the present disclosure.

[0026] Figure 10 This is a schematic diagram comparing the average and standard IGD values ​​of a wireless network optimization method according to one embodiment of the present disclosure.

[0027] Figure 11 This is a schematic diagram comparing the average and standard HV values ​​of a wireless network optimization method according to one embodiment of the present disclosure.

[0028] Figure 12 This is a schematic structural block diagram of a wireless network optimization device according to one embodiment of the present disclosure.

[0029] Figure 13 This is a schematic structural block diagram of an electronic device according to one embodiment of the present disclosure. Detailed Implementation

[0030] The present disclosure will now be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0031] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0032] In forest fire monitoring, sensor nodes often operate in extreme environments characterized by low normal traffic but sudden high loads. When a fire breaks out, transmission power must be increased immediately to ensure high-definition image transmission. However, if the usual low-power strategy is maintained, critical data will be lost due to channel congestion or insufficient power. Conversely, if high-power standby is maintained for an extended period, the battery will be depleted within months of routine monitoring, leaving the sensor nodes unable to function when the fire actually breaks out.

[0033] To address this, this disclosure proposes a wireless network optimization method that achieves refined perception of network status by extracting features from the state data of sensor nodes. Based on the extracted state feature vectors, a multi-objective utility function balancing throughput and energy consumption is constructed. The optimal value of this utility function is then used to generate optimization strategies for sensor nodes adapted to the current scenario. In the event of a fire, this can guide sensor nodes to increase their transmission power to ensure the transmission of critical data; while during routine monitoring, it enables sensor nodes to reduce power consumption and enter an energy-saving mode.

[0034] To facilitate description and make the technical solutions of this disclosure easier to understand, the terminology of this disclosure will be explained before describing the technical solutions of this disclosure.

[0035] Spiking neural networks (SNNs) are distributed neuron models that use discrete pulse sequences to transmit information and achieve local computation and learning based on event-driven and temporal plasticity. They can extract spatiotemporal features and make online decisions with low power consumption.

[0036] Neuron nodes are the basic computational units in spiking neural networks. They are used to simulate the information processing mechanism of biological neurons and can capture and transmit dynamic features and causal relationships in input data in a precise time-encoded manner.

[0037] The wireless network optimization method disclosed herein is applicable to large-scale self-organizing edge IoT systems. Its core computing and decision-making processes are distributed across various terminal sensing devices, rather than relying on a central server. For example, in a field environmental monitoring network, multiple independently powered sensor nodes (such as temperature, humidity, or gas sensors) are scattered throughout a forest or mountainous area, each integrating a low-power microcontroller. All state perception, pulse coding, feature extraction, and optimization strategy solving are completed locally at the sensor nodes. They only exchange pulse signals with other sensor nodes via wireless channels to obtain neighborhood information, eliminating the need to upload raw data to a remote server for centralized processing. This significantly reduces communication energy consumption and end-to-end latency. Even if some sensor nodes fail, the remaining sensor nodes can still autonomously adjust their optimization strategies based on local information, achieving seamless switching between different task modes such as emergency high-throughput or long-term low-power operation. This aligns with the application requirements of edge devices with limited resources but needing intelligent autonomy.

[0038] Figure 1 A schematic diagram illustrating the overall flow of a wireless network optimization method according to one embodiment of this disclosure is shown. Figure 1 The method M100 shown includes steps S110 to S150. This method is based on a distributed spiking neural network for wireless network optimization and can be performed by a base station, edge server, IoT gateway, or terminal device (such as a mobile phone, sensor, or drone).

[0039] In step S110, the status data of each sensor node in the wireless network is obtained. Each sensor node corresponds to a neuron node of the distributed spiking neural network. The status data is used to describe the communication capability, energy reserve and location information of the sensor node in the wireless network.

[0040] During the operation of the wireless network, the status data of each sensor node is collected in real time. Each sensor node is associated with a neuron node in its corresponding distributed spiking neural network. An isomorphic relationship is established between the physical network topology and the artificial neural network structure, allowing the overall situation of the wireless network to be modeled and evolved through a set of cooperating neuron nodes.

[0041] Preferably, the Distributed Spike Neural Network (DSNN) consists of several physically separated neuron nodes. Each neuron node independently performs optimization updates based only on its local state and the spiking events of its neighboring neuron nodes. It achieves global decision-making through lightweight signaling coordination, enabling adaptive learning and online optimization without centralized control of the neuron nodes.

[0042] Preferably, the wireless network is a multi-hop wireless network, consisting of multiple sensor nodes with wireless transceiver capabilities, which transmit and receive signals through a shared wireless channel. The wireless network can be any of the following in the Internet of Things: wireless sensor network, cellular network, Wi-Fi LAN, or Bluetooth Personal Area Network.

[0043] Preferably, a sensor node is a basic functional unit in a wireless network that has sensing, communication, and computing capabilities and is responsible for collecting environmental or device status data in real time.

[0044] Preferably, the status data is a set of key parameters used to characterize the real-time operating status and environmental conditions of the sensor node in the wireless network, including the sensor node's communication capabilities (such as channel quality, signal strength and / or signal interference level), energy reserves (such as remaining battery power and / or energy level), and location information.

[0045] Preferably, the neuron nodes in the distributed spiking neural network are constructed based on the Leaky Integrate-and-Fire (LIF) spiking neuron model.

[0046] In step S120, feature extraction is performed on the state data of the sensor node to determine the state feature vector of the sensor node.

[0047] Directly using the state data of sensor nodes for optimization decisions is susceptible to noise interference and struggles to handle nonlinear relationships. However, by performing nonlinear mapping and spatiotemporal information fusion on the state data collected by each sensor node, key state features reflecting the operational status of the wireless network can be extracted and encoded into structured state feature vectors. This approach not only aligns with bio-inspired low-power computing paradigms but also effectively captures the spatiotemporal correlations and causal patterns in dynamic wireless networks.

[0048] Preferably, the state feature vector is a low-dimensional and compact numerical representation. The state feature vector is no longer limited to the direct reading of a single physical quantity, but contains the correlation between multi-dimensional state features, dynamic evolution trends, and comprehensive situational information in the current wireless network environment.

[0049] In step S130, a utility function is constructed based on the state feature vector of the sensor node, with the optimization objectives of maximizing wireless network throughput and minimizing sensor node energy consumption.

[0050] The essence of wireless networks is a multi-agent collaborative system under resource constraints, which must strike a balance between performance and overhead. Directly setting fixed rules (such as reducing the transmission power of sensor nodes when the battery level is below 30%) is too rigid. However, by constructing parameterized utility functions, it is possible to quantify the trade-offs between multiple objectives.

[0051] Preferably, a utility function is constructed based on the state feature vector of the sensor node and combined with the state feature vectors of other sensor nodes in the wireless network.

[0052] In one specific embodiment, the utility function is: in, Represents the utility function. This represents the combination of transmit power from other sensor nodes. Indicates noise power. This represents the state feature vector of the sensor node, where α and β are weighting coefficients, and the first parameter in the utility function represents the throughput gain. The first parameter represents the channel quality feature in the state feature vector; the second parameter represents the energy cost. The energy characteristics of the sensor nodes in the state feature vector. This indicates the transmit power of the sensor node. This represents the sum of interference from other sensor nodes.

[0053] In step S140, the optimization strategy for the sensor node is determined by maximizing the utility function. The optimization strategy is used to adjust the resource configuration of the sensor node.

[0054] By maximizing the utility function, resource consumption can be minimized while meeting specific performance metrics. This not only improves the efficiency of individual sensor nodes but also enhances the stability of the wireless network, enabling it to extend coverage. Furthermore, considering the complexity and variability of real-world application scenarios, the utility function-based optimization method exhibits strong adaptability and flexibility, allowing for dynamic adjustments to optimization strategies based on actual conditions, thus better addressing various challenges.

[0055] Preferably, by maximizing the utility function, an optimization strategy that maximizes the individual benefits of the sensor nodes is determined. This optimization strategy, as a decision output, is used to guide the sensor nodes to dynamically adjust their operating parameters, such as transmit power or communication mode. The optimization strategy includes the sensor node's transmit power, routing selection, and sleep / wake-up strategy.

[0056] Preferably, the optimization strategy for each sensor node in the wireless network is determined based on Nash equalization.

[0057] In step S150, the wireless network is optimized based on the optimization strategy of the sensor nodes.

[0058] Based on the optimization strategy, the resource configuration and communication behavior of sensor nodes are adjusted in real time. These local adjustments generate a superposition effect through wireless signal propagation and interaction with neighboring sensor nodes, gradually forming a coordinated and consistent operating mode within the wireless network range, thereby achieving the overall performance optimization goal of increased throughput, reduced energy consumption and enhanced connection stability.

[0059] This disclosed wireless network optimization method achieves adaptive resource optimization based on a distributed spiking neural network in multi-hop wireless networks. Specifically, it generates a state feature vector containing spatiotemporal correlations by nonlinearly extracting features from the state data mapping of sensor nodes. Based on this state feature vector, a utility function balancing throughput gain and energy consumption cost is constructed. This utility function is then maximized through Nash equilibrium, autonomously determining optimization strategies such as optimal transmit power, routing selection, and sleep policies. No central control is required, achieving global coordination while ensuring the individual performance of sensor nodes. It can dynamically balance performance and energy efficiency in different application scenarios, improving wireless network throughput, reducing overall energy consumption, and enhancing connection stability, effectively extending the wireless network lifetime. This achieves decentralized, low-overhead, and highly robust wireless network optimization.

[0060] Regarding step S120, which involves extracting features from the state data of the sensor node to determine the state feature vector of the sensor node, in some embodiments of this disclosure, it may include, for example... Figure 2 Steps S1201 to S1202 are shown.

[0061] In step S1201, pulse coding is performed on the state data of the sensor node to determine the pulse sequence of the sensor node.

[0062] The multidimensional state data collected by sensor nodes is processed using pulse coding models (such as rate coding, integral coding, or time coding) to map it into a set of pulse sequences with specific firing patterns.

[0063] Preferably, the following rate coding mechanism is used to pulse-code the state data of the sensor nodes: in, s i ( t ) represents the neuron node corresponding to the sensor node. i pulse sequence, f i This represents the spiking rate of the neuron node corresponding to the sensor node. t i k This represents the firing time of the k-th pulse of the neuron corresponding to the sensor node. δ ( t - t i k ) represents the Dirac function, N i For time windows T The number of pulses in the neuron node corresponding to the internal sensor node. x i For characteristic values ​​of state data (such as channel quality), the pulse firing rate is positively correlated with the characteristic value.

[0064] In step S1202, feature extraction is performed on the pulse sequence of the sensor node to determine the state feature vector of the sensor node.

[0065] The pulse sequences of sensor nodes are subjected to nonlinear spatiotemporal aggregation and dynamic filtering to explore the coupling relationship and evolution law between different state dimensions, and generate a low-dimensional structured state feature vector.

[0066] By pulse coding the state data of sensor nodes and employing a rate coding mechanism to convert it into pulse sequences with temporal dynamic characteristics, the process of mapping continuous state values ​​to brain-like event signals is realized, which can intuitively reflect the intensity of changes in the operating state of sensor nodes. The generated pulse sequences are then subjected to nonlinear spatiotemporal aggregation and dynamic filtering to extract the coupling relationships and temporal evolution patterns between multidimensional states, generating low-dimensional, compact, and semantically rich state feature vectors. This approach not only preserves the temporal dynamic characteristics of the original data but also significantly reduces computational and communication overhead through event-driven sparse representation, while enhancing the ability to perceive complex network situations and improve noise robustness, effectively improving the adaptive optimization performance of wireless networks in dynamic environments.

[0067] Regarding step S1201, pulse encoding is performed on the state data of the sensor node to determine the pulse sequence of the sensor node. In some embodiments of this disclosure, this may include, for example... Figure 3 Steps S310 to S320 are shown.

[0068] In step S310, the optimization priorities corresponding to the throughput and energy consumption of the sensor nodes are obtained respectively.

[0069] Wireless networks face a dynamic, heterogeneous, and resource-constrained environment. Using fixed optimization weights will fail to address the complex and ever-changing real-world demands. Therefore, considering the local network environment and service requirements of each sensor node, the degree of focus on the two key objectives of increasing throughput and reducing energy consumption at the current moment is determined and quantified into two adjustable parameters: throughput optimization priority and energy consumption optimization priority. These priorities are used in the corresponding terms of the weighted utility function, thereby influencing the direction of the final optimization strategy.

[0070] In step S320, based on the mapping relationship between optimization priority and pulse parameters, pulse coding is performed on the state data of the sensor node to determine the pulse sequence of the sensor node. The pulse parameters are used to determine the pulse firing frequency and pulse relative phase of the pulse coding.

[0071] By prioritizing throughput and energy consumption, pulse parameters—namely, pulse firing frequency and pulse relative phase—in the pulse coding model are adjusted as control signals. Optimizing the mapping between priority and pulse parameters allows state data to be encoded into pulse sequences of different patterns, thus influencing the biases in subsequent feature extraction and decision-making processes. For example, prioritizing high throughput increases the pulse frequency of corresponding key state dimensions to enhance their response weight in wireless networks.

[0072] By acquiring the optimization priorities corresponding to the throughput and energy consumption of sensor nodes respectively, and performing pulse coding on the state data based on the mapping relationship between these priorities and pulse parameters, a closed-loop control from high-level optimization intent to low-level perception representation is achieved. The pulse firing frequency and relative phase are dynamically adjusted according to the current network environment and task requirements, enabling key state dimensions to generate more responsive pulse sequences under high-priority objectives, thereby obtaining stronger representation weights in feature extraction and decision-making processes. This not only breaks the static mode of traditional pulse coding but also introduces a brain-like attention mechanism, allowing wireless networks to focus on communication performance or energy efficiency on demand. This improves the flexibility, timeliness, and context adaptability of distributed optimization, ultimately achieving intelligent resource collaboration and adaptive operation for complex dynamic scenarios.

[0073] Regarding step S150, optimizing the wireless network based on the sensor node optimization strategy may include, in some embodiments of this disclosure, the following: Figure 4 Steps S1501 to S1502 are shown.

[0074] In step S1501, the optimization strategy is decoded based on the target pulse emission frequency and the target time interval to generate the sensor node's transmit power, routing selection, and sleep / wake-up strategy. The routing selection is used to determine the path for the sensor node to transmit to the next sensor node in the wireless network, and the sleep / wake-up strategy is used to determine the working state of the sensor node.

[0075] Based on the target pulse firing frequency determined by maximizing the utility function and the target time interval between adjacent neuron pulses, specific control parameters suitable for wireless communication scenarios are derived in reverse. These parameters include transmit power, data forwarding path selection (i.e., routing), and sleep / wake-up strategies, enabling the optimization results in the pulse domain to be implemented as resource scheduling actions in real networks.

[0076] For example, the target pulse emission frequency of neuron nodes above a frequency threshold is mapped to the high transmission power of the corresponding sensor node, indicating that the sensor node should enhance signal strength to ensure communication quality. The target pulse emission frequency of neuron nodes below a frequency threshold is mapped to the energy-saving or low-power mode of the sensor node. By analyzing the relative phase difference of pulses between neighboring channels in different directions, i.e., the consistency and leading edge of the target time interval, the priority path for data forwarding is determined, and the one with the leading phase is selected as the next-hop relay node, i.e., the next sensor node, thus completing the routing selection. At the same time, the node activity level is judged based on the length of continuous pulse-free time. If the silence period exceeds a threshold, a sleep-wake strategy is generated to control the sensor node to enter a low-power sleep state and automatically wake it up at a preset time or when an event is triggered, thereby realizing the comprehensive decoding and configuration of the sensor node's transmission power, routing behavior, and working mode.

[0077] In step S1502, the wireless network is optimized based on the sensor node's transmit power, routing selection, and sleep / wake-up strategy.

[0078] Based on the transmit power, routing selection, and sleep / wake-up strategies of sensor nodes, the wireless communication module, data forwarding mechanism, and power management unit of each sensor node are configured in real time. These local adjustments propagate collaboratively among multiple sensor nodes, creating a cumulative effect, thereby achieving comprehensive performance optimization goals such as increased throughput, reduced energy consumption, improved connection stability, and extended lifespan across the entire network.

[0079] The optimization strategy is decoded based on the target pulse emission frequency and target time interval, transforming the pulse pattern output by neuromorphic computing into specific transmission power, routing selection, and sleep / wake-up strategies, achieving a precise mapping from neuromorphic signals to physical network control. Based on these strategies, the communication, forwarding, and power modules of each sensor node are dynamically configured, enabling on-demand and prioritized allocation of network resources. By fully leveraging the advantages of pulse coding in temporal representation and multi-dimensional information fusion, the reliability of data transmission in critical areas is improved, and overall energy consumption is significantly reduced through intelligent sleep and efficient routing. Ultimately, self-organization and adaptive optimization of the wireless network are achieved without centralized scheduling.

[0080] Regarding step S150, optimizing the wireless network based on the sensor node optimization strategy, in some embodiments of this disclosure, may further include, for example... Figure 5 Steps S510 to S520 are shown.

[0081] In step S510, the wireless network is optimized based on the optimization strategy of the sensor nodes to determine the new state data of the sensor nodes.

[0082] After the sensor node-based optimization strategy completes one round of resource allocation, the sensor nodes enter a new operating state. At this point, it is necessary to re-collect new state data from the sensor nodes reflecting the current wireless network situation for the next round of pulse coding and utility function updates, thereby supporting continuous optimization and online learning in dynamic environments.

[0083] In step S520, the wireless network is iteratively optimized based on the new state data of the sensor node until the difference between the optimization strategy of the sensor node and the optimization strategy of the sensor node in the previous iteration is less than the target threshold in a continuous number of iterations.

[0084] Feature extraction is performed on the new state data of sensor nodes collected after the previous optimization strategy is executed, and a new round of utility function construction and optimization strategy solution is initiated. By comparing the differences in strategy output (such as changes in transmit power) between consecutive iterations, it is determined whether the wireless network has reached stability. When the difference between the optimization strategy of each sensor node and its previous optimization strategy is less than a preset threshold in several consecutive iterations, the wireless network is considered to have reached Nash equilibrium or local steady state, and further updates are stopped, thus completing a full adaptive optimization cycle.

[0085] Based on optimization strategies, the system allocates resources for the wireless network and acquires new state data from sensor nodes, achieving closed-loop feedback for decision execution and perception. This enables real-time capture of network changes after strategy implementation. A new round of optimization is initiated using the new state data. Through continuous iteration and using a convergence criterion that the difference between strategies across multiple rounds is less than a target threshold, the entire wireless network gradually approaches a stable operating state. This system not only supports online tracking and adaptive adjustment of dynamic environments, effectively addressing interference from channel fluctuations, energy attenuation, and topology changes, but also avoids strategy oscillations and overcomputation. While ensuring communication performance, it significantly reduces system overhead, ultimately achieving autonomous convergence, intelligent evolution, and efficient and stable operation of the wireless network in complex scenarios.

[0086] In one specific embodiment, after each iteration of the wireless network, it may further include, as follows: Figure 6 Steps S610 to S620 are shown.

[0087] In step S610, the gradient of the utility function of the sensor node with respect to the optimization strategy of the sensor node is calculated. The gradient represents the trend of change in throughput gain and energy consumption cost under the optimization strategy.

[0088] Relying solely on enumeration or heuristic rules for policy selection is insufficient to handle high-dimensional, nonlinear, and time-varying wireless environments. However, by calculating the gradient of the utility function, sensor nodes can autonomously find the optimal path in complex multi-objective trade-off spaces. This not only enables them to make decisions autonomously but also allows them to understand the causal relationships behind those decisions, thus creating a cognitive wireless network.

[0089] In step S620, the synaptic weights of the neuron nodes corresponding to the sensor nodes are updated by backpropagation based on the gradient.

[0090] By using the gradient as a performance feedback signal and combining it with the pulse time-dependent plasticity algorithm (STBP) of the spiking neural network, the gradient information is backpropagated to the corresponding neuron node in the distributed spiking neural network and used to update its synaptic weights, so that the wireless network can gradually enhance the state behavior mapping relationship that is conducive to improving utility.

[0091] By calculating the gradient of the sensor node utility function relative to the optimization strategy, the changing trend between throughput gain and energy cost resulting from adjusting resource allocation in the current state is accurately quantified, providing clear directional guidance for distributed decision-making. Based on this gradient information, the synaptic weights of neuron nodes are updated via backpropagation. Combined with event-driven learning mechanisms such as the impulse time-dependent plasticity algorithm, online adaptive optimization of neuron nodes is achieved. This not only enables immediate decision-making but also allows continuous learning from historical experience, gradually strengthening state-behavior mapping relationships that improve overall performance, significantly enhancing the adaptability of wireless networks to dynamic environments and their long-term operational stability.

[0092] The following will use specific implementation examples to further explain the wireless network optimization method disclosed herein.

[0093] The wireless network optimization method disclosed herein employs a distributed spiking neural network as its core architecture. It utilizes the dynamic characteristics of LIF spiking neuron models (i.e., neuron nodes) to extract time-varying and topological features of the network environment, and generates distributed resource scheduling strategies through a fully connected layer constructed from the spiking neuron model. Specifically, the LIF spiking neuron model input layer receives real-time state data of the multi-hop wireless network after rasterization processing, converting state data such as node location, channel quality, remaining energy, and interference intensity into pulse sequences. In the encoding stage, a cluster of neurons based on the backpropagation (STBP) mechanism of the spiking neural network extracts features from the input pulse sequences, capturing the spatial structure and dynamic changes in channel quality of the communication links between sensor nodes. In the decoding stage, the fully connected spiking neuron layer, based on the state feature vector after pulse encoding, generates the optimal transmit power, routing selection, and sleep / wake-up strategy results for each node by adjusting the pulse firing frequency and time interval.

[0094] Leveraging the temporal information processing advantages of spiking neural networks (SNNNs) and the decision-making capabilities of Nash equilibrium, a DSNN-NES fusion architecture is constructed to address the multi-objective dynamic optimization requirements of complex wireless networks. Specifically, the spiking neural network uses a spatiotemporal coding mechanism to transform the temporal changes in sensor node states (such as channel quality, remaining energy, and interference levels) in multi-hop wireless networks into pulse sequence inputs. A cluster of neurons based on the backpropagation mechanism of the spiking neural network captures the dynamic evolution of the network topology and the coupling characteristics between nodes. The distributed spiking neural network utilizes the dynamic accumulation of neuronal membrane potentials and threshold firing characteristics to achieve real-time response and feature extraction for transient events such as link quality mutations and user access / departure. Based on this, the Nash equilibrium strategy maps the extracted state feature vectors to utility function parameters in a multi-user non-cooperative game. By iteratively solving for the optimal reaction function (optimization strategy) of each sensor node, a dynamic balance between throughput and energy consumption is achieved in distributed decision-making. This fusion architecture leverages the advantages of distributed spiking neural networks in time-varying environment perception, low-power parallel processing, and efficient encoding of sparse information, while also utilizing the mathematical rigor of the Nash equalization strategy to handle the temporal dependence of the decision-making process, significantly improving the multi-objective optimization accuracy and real-time performance of complex wireless networks in dynamic scenarios.

[0095] Specifically, in multi-hop wireless networks, due to the dispersed distribution of sensor nodes, dynamic changes in links, and multi-user contention, resource scheduling must simultaneously consider data transmission efficiency and energy utilization cost. Simply pursuing a single objective optimization is insufficient to meet the actual needs of complex scenarios. To address this issue, a mathematical model for a multi-objective optimization problem is constructed, with the core objective being to simultaneously maximize total throughput and minimize total energy consumption.

[0096] Maximizing total throughput involves making full use of network link resources to maximize the sum of effective data transmission rates for all source-to-destination sensor node pairs. Specifically, this requires considering link capacity limitations on each transmission path, the attenuation effect of inter-node interference on the rate, and the success rate of data packet delivery during multi-hop forwarding, ultimately maximizing the total amount of data that the wireless network can transmit per unit time.

[0097] Minimizing total energy consumption involves reducing the sum of energy consumption of all sensor nodes in the network, including the transmit energy consumption, receive energy consumption, and circuit energy consumption during relay forwarding. This is achieved by optimizing transmission power allocation, rationally selecting forwarding paths to reduce hop count, and dynamically adjusting the sleep state of sensor nodes. This ensures data transmission requirements are met while avoiding unnecessary energy waste and extending the overall lifespan of the wireless network.

[0098] Among them, consider a containing n A multi-hop wireless network for each user (i.e., sensor node), each user i With transmission power p i The goal of data transmission is to achieve a trade-off between maximizing throughput and minimizing energy consumption.

[0099] user i The utility function must simultaneously reflect throughput gains and energy costs, and is therefore defined as: (1) The first parameter is the user. i Throughput (based on Shannon's formula, unit: bit / s). h i For channel gain, σ 2 For noise power, The first parameter is the sum of interference from other users; the second parameter is the energy cost. c i >0 represents the energy consumption coefficient (reflecting the energy consumption cost per unit of power). p i For users i The transmit power (i.e., the strategy variable). This indicates the power policy combinations of other users.

[0100] Since throughput and energy consumption are conflicting objectives, there is no optimal solution to simultaneously optimize both objectives. Given that an optimal solution does not exist, the challenge in studying this problem is determining the type of solution to seek. For this type of multi-objective optimization problem, it is difficult to directly find a Pareto optimal solution. Therefore, while attempting to find a set of solutions that simultaneously improves both objectives is impractical, improving one objective without negatively impacting the other is possible; this is what we call a weak Pareto optimal solution.

[0101] Furthermore, in game theory, Nash equilibrium is a core concept for measuring the stable state of a multi-player non-cooperative game. Its core idea is: in a game containing... n In a game involving multiple participants (e.g., sensor nodes), if each participant chooses a strategy, and no participant can gain a higher payoff by unilaterally changing their own strategy while the strategies of other participants remain unchanged, then this strategy combination is called Nash equilibrium.

[0102] Nash equilibrium is not only applicable to finite-policy games but can also be extended to infinite-policy spaces (such as continuous variable optimization). In wireless communication, Nash equilibrium is often used to model multi-user resource contention problems (such as power control and channel allocation), where each user's policy is a continuous variable (such as transmit power), and the utility function needs to be defined in conjunction with the communication scenario (such as throughput and energy consumption). Assume the set of game participants is... Participants i The strategy set is P i Strategy combination Represents the strategy choices of all participants (where ). Indicates participants i In strategy combination p The utility function under the Nash power balancing strategy It must satisfy the following condition: Each user cannot increase their own utility by individually adjusting power, assuming other users' policies remain unchanged. .

[0103] By analyzing the utility function v i about p i Taking the derivative and setting it to zero, we can obtain the user's... i The optimal response function (the optimal power of itself given the power of other users): (2) in, Indicates a non-negativity power constraint. Nash equalization. p It is the intersection of all reaction functions (i.e. For all i (Established).

[0104] Nash equalization incorporates throughput and energy consumption into a unified utility function, enabling each user to autonomously weigh transmission gains against energy costs in distributed decision-making, ultimately converging to a globally stable policy combination. This mechanism requires no centralized control, is suitable for large-scale wireless networks, and is particularly effective in multi-hop scenarios by dynamically adapting to link interference and node energy changes, achieving efficient and energy-saving resource allocation.

[0105] In addition, the state data of sensor nodes in the wireless network (such as node power, channel gain, and interference intensity) are converted into pulse sequences and a rate coding mechanism is used. (3) (4) in, s i ( t ) represents the neuron node corresponding to the sensor node. i pulse sequence, f i This represents the spiking rate of the neuron node corresponding to the sensor node. t i k This represents the firing time of the k-th pulse of the neuron corresponding to the sensor node. δ ( t - t i k ) represents the Dirac function, N i For time windows T The number of pulses in the neuron node corresponding to the internal sensor node. x i For characteristic values ​​of state data (such as channel quality), the pulse firing rate is positively correlated with the characteristic value.

[0106] For presynaptic neurons of neuronal nodes j and postsynaptic neurons i ,like t i (Postsynaptic neuron firing time) t j (Presynaptic neuron firing time), the synaptic weights of this neuron node are updated as follows: (5) in, This represents the update amount of synaptic weights (positive values ​​indicate weight enhancement, negative values ​​indicate weight reduction). η The learning rate controls the magnitude of weight updates, typically a positive number between 0.01 and 0.1 to avoid excessively large update magnitudes that could lead to model instability. It represents the time difference between the firing of neuronal impulses before and after the activation (absolute value, reflecting the temporal correlation between the activation of the two). This is the time constant (which controls the decay rate of the time difference's influence on the weight; it is usually set to 10~50ms. The larger the value, the wider the range of influence of the time difference). For symbolic functions ( t i < t j That is, the postneuron fires before the preneuron, returning +1; if t i > t j That is, the first neuron fires and returns -1. Conversely, the weights decay. This rule captures the spatiotemporal correlation features of the wireless network state, and finally outputs a state feature vector. ( M (Dimension of state features).

[0107] Based on state feature vectors q Define the user (i.e., the sensor node). k The utility function is: (6) in, Represents the utility function. This represents the combination of transmit power from other sensor nodes. Indicates noise power. This represents the state feature vector of the sensor node, where α and β are weighting coefficients, and the first parameter in the utility function represents the throughput gain. The first parameter represents the channel quality feature in the state feature vector; the second parameter represents the energy cost. The energy characteristics of the sensor nodes in the state feature vector. This indicates the transmit power of the sensor node. This represents the sum of interference from other sensor nodes.

[0108] Each user finds the optimal power by maximizing their own utility. In other words, optimization strategy: (7) By combining the optimal power of all users, we obtain the Nash equilibrium solution. ,satisfy: This means that Nash equilibrium is achieved.

[0109] like Figure 7 As shown, the dynamic characteristics of sensor nodes in a wireless network are captured through the pulse coding mechanism of neurons in a distributed spiking neural network. Combined with the game theory framework of Nash equilibrium, distributed decision-making is achieved, forming a closed-loop interactive mechanism of feature extraction, decision optimization, and feedback iteration. This provides quantifiable theoretical support for multi-objective optimization of complex wireless networks. The output of the neurons in the distributed spiking neural network is a spatiotemporally encoded state feature vector of the sensor nodes in the wireless network. q This state feature vector contains two key types of information: channel quality features. (such as the dynamic changes in link gain and interference intensity) and node energy characteristics (Such as the temporal distribution of remaining power and energy consumption rate), these features are generated by encoding the firing frequency and time-related patterns of neuron nodes based on the state data of sensor nodes. The solution of Nash equalization is the optimal transmit power of each sensor node. p The data is transformed into new state data and re-inputted into the input layer of the corresponding neuron node. When the sensor node adjusts its transmission power, physical quantities such as channel interference intensity and energy consumption rate change. After rasterization, these changes are transformed into a new pulse sequence, driving the entire distributed spiking neural network to perform the next round of feature updates.

[0110] like Figure 8 As shown, during the initialization phase, the neurons in the distributed spiking neural network receive initial state data (node ​​location, initial energy, etc.) from the sensor nodes in the wireless network. Through the dynamic accumulation of membrane potential in the LIF neuron node model, the initial topological correlation between sensor nodes (such as the impact of physical distance on channel quality) is captured, and the initial state feature vector is output. q 0. The Nash load balancing module is based on q 0. Construct a utility function (i.e., formula (6)), where the throughput revenue term α · f ( , p Energy consumption cost item β · g ( , p The weighting coefficients of ) α =0.6、 β =0.4) reflects the priority of the optimization objective. Each sensor node solves for the optimal response (i.e., the optimization strategy) by maximizing its own utility (i.e., formula (7)). p 0. Initial Equilibrium Strategy p After 0 is applied to the wireless network, each sensor node generates new state data (such as changes in node energy consumption and channel quality fluctuations), which are input into the corresponding neuron node model through pulse coding (Equations (3) and (4)) to trigger the secondary activation of the neuron cluster. At this time, the backpropagation mechanism strengthens the learning of the correlation between policy adjustment and state change through synaptic weight adjustment (i.e., Equation (5)) to generate an updated state feature vector. q 1. The optimization strategy iteration includes: combining Nash equalization with the updated state feature vector. q 1. Recalculate the utility function of the sensor nodes, and solve for the optimal response of each sensor node in the new state feature space. p 1. Due to the updated state feature vector q 1 already includes information on the impact of the previous optimization strategy, so p 1 will be more p 0 is closer to the global optimization objective. When the difference between two consecutive equilibrium solutions (i.e., || p k - p k-1 ||) is less than the preset threshold, and the output state feature vector q k The iteration terminates when the signal tends to stabilize (i.e., the pulse firing pattern of the neuron nodes does not change significantly). p k This is the globally optimal solution that adapts to the dynamic characteristics of wireless networks.

[0111] In addition, to comprehensively evaluate the performance of the method disclosed herein, experimental verification was conducted on a multi-objective optimization function for multi-hop wireless communication networks. The performance of the method disclosed herein (DSNN-NES), the improved multi-objective ant search optimizer (IMOAAO), the dual population constraint multi-objective optimization algorithm with crossover and mutation operators (DCCMO), and the MOSAIC algorithm based on deep neural network (DNN) were compared. Figure 9 For visualization of the results of each method on this function, such as Figure 9As shown, the red line represents the true Pareto front, the blue square represents the result obtained by the IMOAAO algorithm, the green circle represents the result obtained by the DCCMO algorithm, the red square represents the result obtained by the MOSAIC algorithm, and the black circle represents the result obtained by the method disclosed in this paper. In the experiment of multi-objective optimization function in multi-hop wireless communication network, the true Pareto front (True PF) represented by the red line is the theoretical optimal boundary for the dual objective optimization of throughput and energy consumption. That is, each point on the front corresponds to the optimal balance state of "it is impossible to reduce energy consumption while increasing throughput", providing an objective benchmark for comparing the performance of various algorithms. By comparing the results distribution of different algorithms, it can be seen that the results of the IMOAAO algorithm (blue square) are mostly concentrated in the "low throughput - medium energy consumption" region, and some points deviate far from the true Pareto front, indicating that it is prone to energy consumption runaway when increasing throughput, and the dual-objective balance effect is poor. Although some results of the DCCMO algorithm (green circle) are close to the front, the overall distribution is relatively scattered, with polarization between "high throughput - high energy consumption" and "low throughput - low energy consumption", failing to form a stable balanced solution set. Although the results of the MOSAIC algorithm (red square) are closer to the front overall, there is still obvious energy consumption redundancy in the "high throughput" region. In contrast, the results of the method disclosed in this paper (black circle) not only closely follow the true Pareto front overall, but also form a continuous and dense optimal solution set in the core region of "low energy consumption - medium-high throughput", which avoids the throughput shortcoming of IMOAAO and overcomes the result dispersion problem of DCCMO, intuitively demonstrating its advantage in dual-objective balance.

[0112] Figure 10 and Figure 11 Detailed experimental results of the method disclosed herein compared with other algorithms on the IGD and HV metrics are presented. Figure 10 and Figure 11 As shown, the method disclosed in this disclosure exhibits significantly better overall performance than the comparative algorithms in experimental tests of multi-objective optimization functions in multi-hop wireless communication networks. From Figure 10 In terms of IGD values, the average IGD value of the method disclosed in this paper is 3.624e-2, which is not only much lower than IMOAAO (7.146e-2) and DCCMO (8.843e-2), but also better than MOSAIC (4.398e-2). This indicates that the average distance between the Pareto optimal solution set obtained by this method and the theoretical optimal frontier is the smallest, and the accuracy of the bi-objective optimization result is the highest. At the same time, the standard IGD value of the method disclosed in this paper is 1.416e-1, which is the lowest among all algorithms, indicating that its solution set stability is stronger under different experimental scenarios, avoiding the performance deviation caused by scenario fluctuations that other algorithms are prone to. Figure 11In terms of HV values, the average HV value of the method disclosed in this paper reaches 8.013e-1, which is significantly higher than that of IMOAAO (1.646e-1), DCCMO (4.356e-1) and MOSAIC (7.149e-1). It also ranks first, which proves that its solution set can cover a wider range of high-throughput and low-energy-consumption regions, and the dual-objective balance effect is better. Although its standard HV value (5.011e-1) is slightly higher than that of MOSAIC (3.125e-1), combined with the higher average HV value, it can be seen that this fluctuation is within a reasonable range based on "covering a better region", and it is still far better than the stability performance of IMOAAO and DCCMO.

[0113] In summary, the method disclosed herein achieves a more theoretically optimal balance between two objectives in multi-hop wireless communication network optimization, while also ensuring good experimental stability, fully verifying its superiority in simultaneously optimizing throughput and energy consumption. Compared to existing algorithms that focus on a single objective or simply weight multiple objectives, this method utilizes a distributed spiking neural network for precise perception of multiple dimensions such as throughput, energy consumption, and interference. Combined with the Nash equalization strategy's ability to coordinate conflicts of interest among multiple users, it can efficiently find a Pareto optimal solution that balances all objectives. For dynamic scenarios in wireless networks such as channel fluctuations, user movement, and topology changes, the event-driven nature of this method allows for real-time capture of network state changes, rapidly transmitting information and updating parameters using pulse signals, avoiding optimization failures caused by information lag in traditional algorithms. Simultaneously, the distributed decision-making mechanism of the Nash equalization strategy enables each sensor node to autonomously adjust its optimization strategy according to the dynamic environment, without relying on centralized control signaling interaction. On the one hand, the distributed architecture of the method disclosed herein does not require centralized computing nodes. Each sensor node can independently complete local information processing and strategy iteration, significantly reducing cross-node signaling overhead. Simultaneously, the lightweight solution logic of the Nash equalization strategy (based on DSNN pulse threshold termination of invalid iterations) reduces computational resource consumption and is adaptable to resource-constrained edge node scenarios. On the other hand, the method disclosed herein does not require large-scale modifications to existing wireless network hardware and can be deployed on 5G / 6G base stations, IoT gateways, and other devices through software upgrades, reducing deployment costs. It also supports rapid adaptation to multiple scenarios, from ultra-dense networks to emergency communication networks, demonstrating significant advantages in practicality and scalability. Furthermore, because the method disclosed herein possesses fault-tolerant characteristics similar to biological neural networks, the failure of a single sensor node will not paralyze the overall optimization system; the tasks of the failed node can be quickly taken over by neighboring nodes. At the same time, the stable equalization characteristics of the Nash equalization strategy can prevent the optimization strategy from collapsing due to malicious user competition for channel resources or sudden network interference.

[0114] Based on any of the above embodiments, this disclosure also provides a wireless network optimization device.

[0115] Figure 12 This is a schematic block diagram of a wireless network optimization device according to one embodiment of the present disclosure.

[0116] like Figure 12 As shown, the wireless network optimization device includes: The data acquisition module 1202 acquires the status data of each sensor node in the wireless network. Each sensor node corresponds to a neuron node in the distributed spiking neural network. The status data is used to describe the communication capability, energy reserve and location information of the sensor node in the wireless network. The feature extraction module 1204 extracts features from the state data of the sensor nodes to determine the state feature vector of the sensor nodes. Function construction module 1206 constructs a utility function based on the state feature vector of the sensor node, with the optimization objectives of maximizing wireless network throughput and minimizing sensor node energy consumption. The strategy determination module 1208 determines the optimization strategy for the sensor nodes by maximizing the utility function. The optimization strategy is used to adjust the resource configuration of the sensor nodes. The network optimization module 1210 optimizes the wireless network based on the optimization strategy of sensor nodes.

[0117] The aforementioned wireless network optimization device can be in the form of computer software, and each module of the aforementioned wireless network optimization device can be implemented through computer software modules.

[0118] The specific implementation process of the functions and roles of each module in the above wireless network optimization device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0119] This disclosure also provides an electronic device corresponding to this. Figure 13 A schematic diagram of the hardware implementation using the processing system is shown.

[0120] The hardware structure of electronic device 1000 can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the hardware. Bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400 such as peripherals, voltage regulators, power management circuits, external antennas, etc. Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, only one connection line is used in this figure, but this does not indicate that there is only one bus or one type of bus.

[0121] For ease of explanation, certain steps of the above method are described in relation to modules. It should be understood that the corresponding module performing one or more steps of the above method may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination thereof.

[0122] This disclosure also provides a readable storage medium storing a computer program that, when executed by a processor, is used to implement the methods described above. A "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of a readable storage medium include: an electrical connection with one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM), etc.

[0123] This disclosure also provides a computer program product, the methods of which can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented wholly or partially as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, all or part of the processes or functions of this disclosure are performed.

[0124] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0125] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, or characteristic described in connection with that embodiment / mode or example is included in at least one embodiment / mode or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0130] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0131] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A wireless network optimization method, characterized in that, Wireless network optimization is implemented based on distributed spiking neural networks. The wireless network optimization method includes: The status data of each sensor node in the wireless network is obtained. Each sensor node corresponds to a neuron node of the distributed spiking neural network. The status data is used to describe the communication capability, energy reserve and location information of the sensor node in the wireless network. Feature extraction is performed on the state data of the sensor node to determine the state feature vector of the sensor node; Based on the state feature vector of the sensor node, a utility function is constructed with the optimization objectives of maximizing wireless network throughput and minimizing sensor node energy consumption. An optimization strategy for the sensor node is determined by maximizing the utility function; this optimization strategy is used to adjust the resource allocation of the sensor node. The wireless network is optimized based on the optimization strategy of the sensor nodes.

2. The wireless network optimization method as described in claim 1, characterized in that, The wireless network is optimized based on the optimization strategy of the sensor nodes, including: The wireless network is optimized based on the optimization strategy of the sensor nodes to determine the new state data of the sensor nodes; Based on the new state data of the sensor node, the wireless network is iteratively optimized until, in a continuous target number of iterations, the difference between the optimization strategy of the sensor node and the optimization strategy of the sensor node in the previous iteration is less than the target threshold.

3. The wireless network optimization method as described in claim 2, characterized in that, Also includes: After each iteration of the wireless network, the gradient of the utility function of the sensor node with respect to the optimization strategy of the sensor node is calculated, and the gradient represents the trend of throughput gain and energy consumption cost under the optimization strategy. Based on the gradient, the synaptic weights of the neuron nodes corresponding to the sensor nodes are updated via backpropagation.

4. The wireless network optimization method as described in claim 1, characterized in that, Feature extraction is performed on the state data of the sensor node to determine the state feature vector of the sensor node, including: The state data of the sensor node is pulse-coded to determine the pulse sequence of the sensor node; Feature extraction is performed on the pulse sequence of the sensor node to determine the state feature vector of the sensor node.

5. The wireless network optimization method as described in claim 4, characterized in that, Pulse coding is performed on the state data of the sensor node to determine the pulse sequence of the sensor node, including: Obtain the optimization priorities corresponding to the throughput and energy consumption of the sensor nodes respectively; Based on the mapping relationship between optimization priority and pulse parameters, pulse coding is performed on the state data of the sensor node to determine the pulse sequence of the sensor node. The pulse parameters are used to determine the pulse firing frequency and pulse relative phase of the pulse coding.

6. The wireless network optimization method as described in claim 1, characterized in that, The optimization strategy for each sensor node in the wireless network is determined based on Nash equilibrium.

7. The wireless network optimization method as described in claim 1, characterized in that, The wireless network is optimized based on the optimization strategy of the sensor nodes, including: Based on the target pulse emission frequency and the target time interval, the optimization strategy is decoded to generate the sensor node's transmit power, routing selection, and sleep / wake-up strategy. The routing selection is used to determine the path for the sensor node to transmit to the next sensor node in the wireless network, and the sleep / wake-up strategy is used to determine the working state of the sensor node. The wireless network is optimized based on the transmit power, routing selection, and sleep / wake-up strategy of the sensor nodes.

8. The wireless network optimization method as described in claim 1, characterized in that, Based on the state feature vectors of the sensor nodes, a utility function is constructed with the optimization objectives of maximizing wireless network throughput and minimizing sensor node energy consumption, including: Based on the state feature vector of the sensor node, and combined with the state feature vectors of other sensor nodes in the wireless network, a utility function is constructed.

9. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes the execution instructions stored in the memory, causing the processor to perform the wireless network optimization method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the wireless network optimization method according to any one of claims 1 to 8.

Images