An event-driven dynamic ordering activation and information fusion method for sensor networks

Abstract

This invention discloses an event-driven dynamic ranking activation and information fusion method for sensor networks. First, a solar generator and battery are integrated into the sensor nodes, and power generation simulation is performed based on a real irradiance dataset. A Poisson model is used to simulate the random occurrence of targets or events to drive the switching of sleep or wake-up states of surrounding nodes. Second, the state ranking value of each node is calculated based on its importance parameters. Joint parameters for the event-driven region and activation ratio in the activation method are designed, and combined with weights driven by link quality, an improved distributed Kalman information fusion method is used to further maximize the efficiency of utilizing the sensor information of all activated nodes. Finally, simulations verify the effectiveness of the designed method. This invention optimizes network performance while maximizing the efficiency of utilizing the sensor information of all activated nodes, further improving sensing accuracy in the sense of minimum variance.

Description

Technical Field

[0001] This invention relates to wireless communication technology, and more specifically to an event-driven method for dynamic sorting activation and information fusion in sensor networks. Background Technology

[0002] Wireless sensor network technology has developed rapidly in recent years and has practical value in many important fields, such as urban management, environmental monitoring, disaster relief, and reconnaissance. It has attracted increasing attention from researchers worldwide and is considered one of the technologies that will have a significant impact on social development in the coming decades.

[0003] Currently, wireless sensor networks (WSNs) are network systems that integrate monitoring, control, and wireless communication functions, and consist of a large number of densely distributed sensor nodes. These nodes typically rely on miniature batteries for power, resulting in very limited energy resources. Due to environmental factors and energy depletion, nodes are prone to failure. Therefore, optimizing the energy efficiency and sensing performance of sensor networks is of great significance.

[0004] Traditional wireless sensor network nodes typically employ a periodic wake-up mechanism, periodically waking up to perform data acquisition and transmission tasks before entering a sleep state to conserve energy. However, this method has several drawbacks. First, the node wake-up cycle is usually fixed and cannot be dynamically adjusted according to real-time needs, leading to energy waste or delayed responses. Second, periodic wake-ups can cause energy imbalances, with some nodes potentially overconsuming energy and running out of power prematurely, thus affecting the normal operation of the entire network. To overcome these problems, self-wake-up technology has been proposed and widely applied in wireless sensor networks. Self-wake-up technology is event-driven, triggering node wake-up by monitoring changes in environmental events. This technology can effectively reduce the energy consumption of wireless sensor networks, extend network lifetime, and improve network response speed and data acquisition efficiency. Therefore, this technology has broad application prospects in the field of wireless sensor networks.

[0005] In wireless sensor networks, each node can independently collect data and transmit and share it with other nodes via wireless communication. However, due to limited node resources and network communication bandwidth, directly sending the raw data collected by all nodes to a central node for processing and analysis is impractical. Furthermore, due to potential node location deviations and measurement errors, directly using data from a single node may lead to inaccurate or unrobust results. To address these issues, distributed data fusion technology has been proposed and widely applied in wireless sensor networks. Distributed data fusion technology allows nodes to perform data fusion and processing within the network to produce more accurate, robust, and reliable results. Currently, most information fusion methods that achieve consensus convergence for fusion errors employ an average consensus strategy, which theoretically requires too many iterations. Therefore, a finite-time control technique is further employed to achieve finite-iteration convergence of fusion errors. Summary of the Invention

[0006] Based on the above analysis and discussion, this invention discloses an event-driven method for dynamic ranking activation and information fusion in sensor networks. First, a solar generator and battery are integrated into the sensor nodes, and power generation simulation is performed based on a real irradiance dataset. A Poisson model is used to simulate the random occurrence of targets or events to drive the switching of sleep or wake-up states of surrounding nodes. Second, the state ranking value of each node is calculated based on its importance parameters. Joint parameters for the event-driven region and activation ratio in the activation method are designed, and combined with weights driven by link quality, improved distributed Kalman information fusion further maximizes the utilization of sensor information from all activated nodes. Finally, simulations verify the effectiveness of the designed method.

[0007] An event-driven method for dynamic sorting activation and information fusion in sensor networks includes the following steps:

[0008] Step 1: Using the Intel Berkeley Labs dataset, deploy a wireless sensor node topology network to form a communication network, then proceed to Step 2.

[0009] Step 2: Integrate the solar generator and battery into the sensor nodes in the topology network, and simulate power generation based on the real irradiance dataset. Use the Poisson model to simulate the random occurrence of the target or event to drive the switching of the sleep or wake-up state of the sensor nodes around the target or event, and then proceed to Step 3.

[0010] Step 3: Model the energy consumption of each sensor node in the topology network to construct the topology network energy model and link quality model. The topology network energy model includes the communication energy consumption model and the sensing energy consumption model. Proceed to Step 4.

[0011] Step 4: Calculate the state ranking value of the sensor node based on the important parameters of the sensor node in the energy model of the topology network and the important parameters of the sensor node in the link quality model. Design the joint parameters of the event-driven region and activation ratio in the ranking activation method. Activate the sensor node in combination with the state ranking value and proceed to step 5.

[0012] Step 5: The awakened sensor nodes perform information fusion using an improved distributed Kalman fusion method, introducing weights driven by link quality. Each sensor node updates its own estimated data based on its own estimated data, neighbor's estimated data, and weight matrix to obtain an accurate target state estimate.

[0013] Compared with the prior art, the advantages of this invention are:

[0014] (1) Unlike the study on the wake-up mechanism of sensor network based on ideal energy constraints, the sorting activation method is extended to the solar-powered self-powered sensor network. System modeling and method improvement are carried out, and the long-term autonomous operation performance of the self-powered network is simulated and analyzed based on real irradiation dataset.

[0015] (2) A joint event-driven and state-level activation method is proposed. Based on a real sensor laboratory dataset, the win-win contribution of this joint mechanism to energy efficiency and sensing performance is verified.

[0016] (3) Compared with traditional information fusion methods, weights driven by link quality are introduced, and the optimal fusion estimation in the sense of minimum variance is achieved through the improved distributed Kalman information fusion. Attached Figure Description

[0017] Figure 1 This is a diagram of a distributed sensor network system, where (a) shows the network topology and working mechanism; and (b) shows the node working status and energy consumption.

[0018] Figure 2 This is the flowchart for the dynamic sorting activation method.

[0019] Figure 3 This is a diagram showing the impact of the event-driven region and activation ratio on perceptual error.

[0020] Figure 4 This is a comparison chart of network lifetimes.

[0021] Figure 5 This is a flowchart of the improved distributed Kalman information fusion method.

[0022] Figure 6 It is a perception trajectory map of specific events after the network has been running for a long time.

[0023] Figure 7This is a comparison chart of average energy consumption and sensing performance.

[0024] Figure 8 Flowchart of an event-driven method for dynamic sorting, activation, and information fusion in sensor networks. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Combination Figures 1 to 8 The present invention provides an event-driven method for dynamic sorting, activation, and information fusion in sensor networks, comprising the following steps:

[0027] Step 1: Node deployment based on the Intel Berkeley Lab dataset is adopted. The wireless sensor network can be represented by a graph G = {ν, ε}, where the node set ν of the graph represents the sensors, and the edge set ε of the graph represents the relationships between sensors, as detailed below:

[0028] By employing node deployment based on the Intel Berkeley Lab dataset, the wireless sensor network is transformed into an undirected, unweighted graph G = {ν, ε}, where the graph node set ν = {1, 2, ..., n} represents the sensors, and the graph edge set... Let ε represent the relationship between sensor i and sensor j. If an edge (i,j)∈ε indicates that information from sensor i can be transmitted to sensor j, but this process is not necessarily reversible. Figure 1 (a) shows a simplified topology diagram of a distributed wireless sensor network deployment. Based on the topology, the adjacency matrix of graph G can be obtained, and DEG is defined. ij Let DEG be the external link count matrix. ij =[b jm ] n×n Let b represent the connectivity of sensor i's neighbor sensor j. If sensor j has a neighbor sensor m, then b jm =b mj =1.

[0029] Step 2: Integrate the solar generator and battery into the sensor nodes in the topology network, and simulate power generation based on the real irradiance dataset (solar irradiance intensity dataset from the 2016 Beijing real-time meteorological data provided by the China National Meteorological Center). Use the Poisson model to simulate the random occurrence of targets or events to drive the switching of the sleep or wake-up state of surrounding nodes, as follows:

[0030] In the area of ​​environmental energy harvesting, devices such as solar generators, vibratory generators, and thermoelectric generators are integrated into wireless sensor network nodes to continuously collect energy from the environment, aiming to achieve long-term autonomous energy operation for the nodes. Therefore, a solar generator and battery are integrated to power the node, with monocrystalline silicon solar panels used for energy replenishment. When the weather is clear, the node continuously collects electrical energy through the solar generator and stores it in the battery for use in subsequent high-power activation modes. Once the battery is fully charged, power generation ceases.

[0031] The power output of a solar generator is mainly affected by the intensity of solar irradiance and the area of ​​the solar panels. The power calculation model for a solar generator is as follows:

[0032]

[0033] P h =P m *A0 (2)

[0034] Among them, P m It is the power density of solar power generation; G tt It is solar irradiance; a, b, and c are all empirical parameters obtained from experiments; P h A0 represents the power output of the solar generator; A0 is the area of ​​the solar panel.

[0035] For most wireless sensor network applications, the targets or events to be monitored are sparse in both time and space. A Poisson model is used to simulate the random occurrence of targets or events, driving the switching of sleep or wake-up states of surrounding nodes. To reduce the average power consumption of the nodes, a low-vibration wake-up switch based on transistor or MEMS technology is integrated into the nodes, ensuring that the nodes are only awakened when a target or event is present in their vicinity. Under this event-driven self-wake-up mechanism, the random occurrence of targets or events is simulated using a Poisson model to drive the switching of sleep or wake-up states of surrounding nodes. Combining the results of the Poisson model simulation of targets or events, and further considering that the known targets or events will be randomly distributed within the deployed two-dimensional space, the nodes in the sensor network can then accurately monitor the target location information.

[0036] Assuming the average number of random events occurring in a network per unit time is λ, the probability function P of r events occurring within a given time unit is:

[0037]

[0038] Where X represents an event.

[0039] In a distributed sensor network, nodes within the event-triggered area are activated and accurately monitor the event target. Assuming the event is randomly distributed in two-dimensional space, it persists for a period of time after its occurrence and then disappears randomly according to the Poisson model. Before the event disappears, its spatial location will randomly shift over time, and the goal of the sensor network is to accurately monitor its trajectory.

[0040] In step 3, the energy consumption of each sensor node in the topology network is modeled to construct a topology network energy model and a link quality model. The topology network energy model includes a communication energy consumption model and a sensing energy consumption model, as detailed below:

[0041] Communication energy consumption is mainly divided into three parts: energy consumption for transmitting and receiving data and energy consumption for data fusion. The energy consumption calculation during the data transmission process of a node depends on the distance between neighbors and the energy consumption per unit bit data packet.

[0042] The communication energy consumption model uses a first-order radio model, where the energy consumption E for transmitting data between adjacent nodes at a distance of d is... transmit (k,d) is:

[0043]

[0044] Among them, E TX Energy consumption per bit of transmitted packet; E fs Amplifier power consumption representing the free-space model; E amp The amplifier power consumption represents the multipath fading model; k is the length of the transmission packet; d0 is the distance threshold, used as a threshold to divide the spatial model.

[0045] The energy consumption E for each node to receive data recieve (k) is:

[0046] E recieve (k)=k×E RX (5)

[0047] Among them, E RX It is the energy consumption per bit of received data packet.

[0048] Energy consumption E of each node during data fusion fusion (k) is:

[0049] E fusion (k)=k×E DA (6)

[0050] Among them, E DA It is the energy consumption per bit of data fusion.

[0051] The energy consumption of a node for data detection and sensing mainly depends on the distance between the node and the target and the required target detection accuracy. When an event occurs, the sensing energy consumption P of each activated node in its vicinity... t for:

[0052]

[0053] Among them, D e It is the accuracy of perception; R d ξ is the distance between the node and the target; ξ, β and μ are all constants.

[0054] In wireless sensor networks, nodes remain awake after activation and are used to sense the location information of targets or events, a process that generates significant sensing power consumption. Additionally, the process of sending, receiving, and fusing the acquired information in the form of data packets also generates significant communication power consumption; at this time, the node is in a high-power state. Nodes do not participate in sensing and communication processes when in sleep mode; therefore, they are in a low-power sleep state, such as... Figure 1 As shown in (b).

[0055] During sensor information transmission, environmental noise, electromagnetic interference, and other factors can compromise transmission reliability, leading to adverse results such as error codes and data packet loss. In the field of communications, the concept of link quality is commonly used to assess this issue. The link quality for data transmission between nodes employs a logarithmic signal path attenuation model, representing the transmission quality by calculating the signal-to-noise ratio (SNR) of the transmitting node at a reference distance d1. The quality calculation model for the link between adjacent transmitting nodes at a distance d is γ. d for:

[0056]

[0057] Among them, P e 1 is the node's transmit power; PL(d1) is the power loss at the reference distance (d1 is the reference distance, which must be determined according to different propagation environments. Since our simulation environment is a small-radius microcell system, the reference distance is set to 1m); η is the path loss exponent (related to the specific indoor environment, generally between 2 and 4); N(0,σ) is a normally distributed random variable with a mean of 0 and a variance of σ. 2 ;P n It is the noise floor.

[0058] Step 4: Calculate the state ranking values ​​of sensor nodes based on the important parameters of sensor nodes in the topology network energy model and the link quality model. Design joint parameters for the event-driven region and activation ratio in the ranking activation method. Activate sensor nodes based on the state ranking values, as follows:

[0059] Step 4.1: Establish a dynamic state sorting mechanism for nodes within the region:

[0060] Event-driven dynamic sorting and activation process of sensor networks, as follows Figure 2 As shown, firstly, the network simulates the occurrence of events based on a Poisson distribution. Due to the self-driven wake-up switch, nodes within a certain distance are temporarily physically awakened after an event occurs. Furthermore, nodes near the event location are first filtered, and some abnormal nodes are removed. At the start of the sorting process, each node sends its important parameters to the fusion center, including remaining energy, distances between nodes, and link quality, which are then maintained and updated at the fusion center.

[0061] Within the event region, the initial state sorting value R for each node is calculated using the following formula. i :

[0062]

[0063] Where, γ ij Distance represents the link quality between node i and its neighbor node j. ij SNE represents the Euclidean distance between node i and its neighbor node j; i is the remaining energy of the node; N is the number of nodes in the entire directed graph network topology.

[0064] To avoid the randomness of ranking calculations for a single sensor, which could reduce robustness, a more stable result is obtained by iterating using the outer link matrix of a node in the following manner. The ranking value of node i is iterated in the (n+1)th iteration.

[0065] Alternative calculation formula as follows:

[0066]

[0067] Among them, DEG ij is the matrix of external links of node i's neighbor node j; α is the damping factor.

[0068] Step 4.2: Design the joint parameters of the event-driven region and activation ratio in the sorting activation method, including the activation region area S and the activation ratio J of sensor nodes within that region, and activate the sensor in combination with the state sorting value.

[0069] Device node:

[0070] Existing sorting methods all require global state sorting of all nodes, which is often uneconomical in terms of energy consumption and information exchange. Based on the physical effect of self-wake-up switches, an event-driven dynamic sorting activation method for sensor networks is proposed. In this state sorting mechanism, nodes within a circular region S surrounding the event can be triggered. Only surviving nodes within region S will perform subsequent sensor sorting, thus avoiding unnecessary activation of nodes too far from the event. Under this driving mechanism, the size of the driving region S is crucial: too large a value for S will fail to effectively reduce the operating cost of subsequent sensor sorting; too small a value for S may cause some potentially valuable nodes to be missed in subsequent sensor ranking. Determining the node activation selection ratio before activating nodes is also critical, improving node transmission accuracy and energy efficiency while ensuring connectivity. The formula for calculating the sensor node selection ratio is as follows:

[0071] J = num / ρS (9)

[0072] Where J is the activation ratio of nodes in the activation area S; ρ is the spatial distribution density of nodes in the topology network; and num is the minimum number of activated nodes required to achieve the specified monitoring accuracy.

[0073] from Figure 3 Simulation results show that although increasing these two parameters means increasing the number of actual activated nodes, they have different effects on the perception accuracy after information fusion. As the activation ratio increases, the sensing error first decreases and then increases, especially when the event triggering region is large. This is because when the activation ratio increases from a relatively small value to a suitable value, more information is aggregated and fused, improving perception performance; however, when the activation ratio increases too much, some nodes with poor perception will also be aggregated, worsening the network's perception performance. The event triggering region is actually physically limited by the self-wake-up switch and cannot be increased indefinitely. According to... Figure 3 The results show that as the event-triggered area increases within its possible range, the sensing error almost always decreases. On the other hand, the network's average energy consumption will undoubtedly increase with the increase of both parameters, as they both lead to the activation of more nodes.

[0074] To highlight the advantages of this invention at the network energy level, the dynamic activation method is compared with the network energy efficiency of random selection activation. Figure 4Under the same conditions, the 10-day performance of this activation method and the random activation method with the same number of nodes was compared. The network running with the random activation method experienced a large number of dead nodes after 5 days, stabilizing at around 20% after 10 days. In contrast, the network running with the designed sequential activation never experienced any dead nodes, demonstrating the method's significant advantage in improving the energy balance of nodes in the network. Furthermore, both methods performed similarly in terms of average remaining energy because they had the same number of active nodes.

[0075] Step 5: The awakened sensor nodes perform information fusion using an improved distributed Kalman fusion method, introducing weights driven by link quality. Each sensor node updates its own estimated data based on its own estimated data, neighbor's estimated data, and the weight matrix to obtain an accurate target state estimate, as detailed below:

[0076] Step 5.1: Based on the logarithmic signal path attenuation model in Step 3, introduce weights driven by link quality and update the weights in real time:

[0077] In the data fusion process of multiple sensors, to obtain the most accurate global information, a communication topology network needs to be established between multiple nodes to enable data interaction between the multiple sensors. In step 3, the link quality calculated by the logarithmic signal path attenuation model represents the ability of nodes to transmit data to each other. Therefore, a weight driven by link quality is introduced, and the weight is updated in real time to reduce the error caused by nodes with relatively poor transmission performance, thereby improving the accuracy of the information fusion process. Therefore, the following definition exists: the original communication topology graph G={ν,ε} is extended to... in, This is the weight matrix.

[0078] Step 5.2: Each sensor node updates its own estimated data based on its own estimated data, neighbor estimated data, and weight matrix:

[0079] The self-estimated data measured by each node, For node i at time k, the system state variable x k The local estimation results are obtained. Each node obtains its neighbor estimation data based on the network topology. This neighbor estimation data represents the self-estimation data of each node in the sensor network with which it has communication relationships. The node updates its own estimation data based on its own, its neighbor estimation data, and the weight matrix set.

[0080] For nodes within the activated region, the linear discrete-time dynamic model of the wireless sensor network is:

[0081] x k+1 =Ax k +Bwk (12)

[0082]

[0083] Where, x k This represents the state at time k; A represents the measurement value of node i at time k;

[0084] B represents the state transition matrix; H represents the noise matrix of the system input; B represents the noise matrix of the system input. i The measurement matrix representing the nodes; w k This represents the process noise at time k. This represents the measurement noise at node i at time k.

[0085] Improved distributed Kalman information fusion method, such as Figure 5 As shown. First, we set an initial measurement information transfer matrix. and a state information transfer matrix As shown in formulas (14) and (15).

[0086]

[0087]

[0088] in, It is the measurement matrix of node i at time k. It is the measurement error covariance matrix.

[0089] Each node aggregates information from all active nodes in the entire network. Considering the impact of link quality on information transmission reliability analyzed in step 3, link quality-driven weights are introduced into the information fusion process, and these weights are updated in real time. For node i, it undergoes iteration d... a times (d) a (where is the maximum number of hops in the network topology), and the final summary results are as follows.

[0090] For node i, the t-th iteration is performed as follows:

[0091]

[0092]

[0093] Furthermore, we also provide the iterative equation for the neighbor j of node i:

[0094]

[0095]

[0096] Where W = [ω ij ]∈R n×n The weight elements in the weight matrix satisfy ω ij =ω ji >0 and ω ii =1; t=1,2,3,...,d a .

[0097] Step 5.3: Each sensor node obtains an accurate target state estimate through prediction and updating, and achieves finite-time convergence of the fusion error:

[0098] In step 5.2, after d a In the next iteration, we obtain the extended graph G = {ν, ε, W} from layer 0 to layer d. a The sensor node sensing information transmission matrix α of the layer i (d a ) and Q i (d a This state variable possesses global information, thus enabling accurate estimation of the target state. During information fusion, the following formulas are used for prediction and updating:

[0099] predict:

[0100]

[0101]

[0102] renew:

[0103]

[0104]

[0105] in, The prior state estimate of node i at time k; Let be the prior estimation error covariance of node i at time k; The posterior state estimate of node i at time k; Let F be the covariance of the posterior estimation error at time k; k-1 Given the system matrix; Y k-1 Let be the process error covariance matrix; T denotes transpose.

[0106] Since the local estimation process of this information fusion method is based on classical local Kalman filtering, the final fusion step only requires combining the local estimation results. Weighted fusion is then performed. Finally, the final fusion result with the minimum variance is obtained by fusing the information referenced by each node. The optimal fusion calculation formula is as follows:

[0107]

[0108] Where, Γ=Σ -1 e(e T Σ -1 e) -1 It is a positive definite matrix. Let n be the cross-covariance matrix, n be the target state dimension, and l be the number of sensors. It is the local process noise covariance matrix of nodes i and j.

[0109] Regarding perception performance, the accuracy of the network's perception of specific event trajectories was first evaluated after long-term operation, such as... Figure 6 As shown in the figure, the event trajectories obtained by distributed information fusion sensing using the proposed activation method almost completely overlap with the results and real trajectories of centralized information fusion sensing. Furthermore, it can be concluded that the excellent sensing performance is a joint contribution of the proposed distributed information fusion algorithm and activation method. In stark contrast, the random activation method, even with the same distributed information fusion, shows a significant decrease in sensing accuracy because many nodes that could have contributed high-quality sensing results die due to early imbalances in network energy states. If only the sensing results of a single nearest node are used without multi-node information fusion, the sensing results of the event trajectories become even worse.

[0110] Regardless of whether activation is random or state-ordered, the perception error of target information is the most important indicator of information fusion. To verify the fusion error of system state information obtained by different wake-up methods, we comprehensively compared the network's energy and perception performance under different operating mechanisms, such as... Figure 7 As shown, the proposed method outperforms distributed information fusion under random activation, and is far superior to single-node sensing, in terms of both average sensing error characterizing sensing performance and average energy consumption characterizing energy performance. This is because the proposed joint activation method can dynamically select the node most suitable for the current sensing task through a comprehensive evaluation of energy and sensing state, while the proposed improved distributed Kalman information fusion method can most effectively utilize the sensing information of all activated nodes.

Claims

1. An event-driven method for dynamic sorting, activation, and information fusion in sensor networks, characterized in that, Includes the following steps: Step 1: Using the Intel Berkeley Labs dataset, deploy a wireless sensor node topology network to form a communication network, then proceed to Step 2. Step 2: Integrate the solar generator and battery into the sensor nodes in the topology network, and simulate power generation based on the real irradiance dataset. Use the Poisson model to simulate the random occurrence of the target or event to drive the switching of the sleep or wake-up state of the sensor nodes around the target or event, and proceed to step 3. Step 3: Model the energy consumption of each sensor node in the topology network to construct the topology network energy model and link quality model. The topology network energy model includes the communication energy consumption model and the sensing energy consumption model. Proceed to Step 4. Step 4: Calculate the state ranking value of the sensor node based on the important parameters of the sensor node in the energy model of the topology network and the important parameters of the sensor node in the link quality model. Design the joint parameters of the event-driven region and activation ratio in the ranking activation method. Activate the sensor node in combination with the state ranking value and proceed to step 5. Step 5: Information fusion is performed among the awakened sensor nodes using an improved distributed Kalman fusion method. Weights driven by link quality are introduced. Each sensor node updates its own estimated data based on its own estimated data, neighbor estimated data, and weight matrix to obtain an accurate target state estimate.

2. The event-driven dynamic sorting activation and information fusion method for sensor networks according to claim 1, characterized in that, In step 1, sensor nodes based on the Intel Berkeley Lab dataset are deployed to transform the wireless sensor network into a graph G = {ν, ε}, where the node set ν represents the sensors and the edge set ε represents the relationships between sensors. The adjacency matrix of graph G is obtained based on the topological network structure, and DEG is defined. ij Let be the outer link matrix, which represents the external connectivity of node i's neighbor node j.

3. The event-driven dynamic sorting activation and information fusion method for sensor networks according to claim 1, characterized in that, In step 2, the solar generator and battery are integrated into the sensor nodes in the topology network, and power generation simulation is performed based on real irradiance datasets. The Poisson model is used to simulate the random occurrence of targets or events to drive the switching of the sleep or wake-up state of surrounding nodes, as detailed below: A solar generator and a battery are integrated to power the sensor node. The solar generator continuously collects electrical energy and stores it in the battery for use by the sensor node in subsequent high-power activation modes. The power calculation model for a solar generator is as follows: P h =P m *A0 (2) Among them, P m It is the power density of solar power generation; G tt It is solar irradiance; a, b, and c are all empirical parameters obtained from experiments; P h A0 represents the power output of the solar generator; A0 is the area of ​​the solar panel. The random occurrence of targets or events is simulated using a Poisson model to drive the switching of sleep or wake-up states of surrounding sensor nodes. The sensor nodes integrate a power supply weak vibration wake-up switch based on transistor or MEMS technology, so that the sensor nodes are only awakened when targets or events appear in the vicinity. Under this event-driven self-wake-up mechanism, the random occurrence of targets or events is simulated using a Poisson model to drive the switching of sleep or wake-up states of surrounding sensor nodes. Combining the simulation of targets or events using the Poisson model, targets or events are randomly distributed in the deployed two-dimensional space. Assuming the average number of random events occurring in a network per unit time is λ, the probability function P of r events occurring within a given time unit is: Where X represents an event.

4. The event-driven dynamic sorting activation and information fusion method for sensor networks according to claim 1, characterized in that, In step 3, the energy consumption of each sensor node in the topology network is modeled to construct a topology network energy model and a link quality model. The topology network energy model includes a communication energy consumption model and a sensing energy consumption model, as detailed below: The communication energy consumption model adopts a first-order radio model, where the energy consumption E for transmitting data between adjacent sensor nodes at a distance of d is... transmit (k,d) is: Among them, E TX Energy consumption per bit of transmitted packet; E fs Amplifier power consumption representing the free-space model; E amp The amplifier power consumption represents the multipath fading model; k is the length of the transmission packet; d0 is the distance threshold, used as a threshold to divide the spatial model. The energy consumption E for each sensor node to receive data recieve (k) is: E recieve (k)=k×E RX (5) Among them, E RX It is the energy consumption per bit of received data packet; Energy consumption E of each sensor node during data fusion fusion (k) is: E fusion (k)=k×E DA (6) Among them, E DA It is the energy consumption per bit of data fusion; The energy consumption of a sensor node for data detection and sensing mainly depends on the distance between the sensor node and the target and the required target detection accuracy. When an event occurs, the sensing energy consumption P of each activated node in its vicinity... t for: Among them, D e It is the accuracy of perception; R d It is the distance between the sensor node and the target; ξ, β, and μ are all constants; The link quality during data transmission between sensor nodes is represented by a logarithmic signal path attenuation model. The transmission quality is indicated by calculating the signal-to-noise ratio (SNR) of each transmitting node at a reference distance d1. The quality calculation model for the link between adjacent transmitting nodes at a distance d is γ. d for: Among them, P e η is the transmit power of the sensor node; PL(d1) is the power loss at the reference distance; η is the path loss exponent; N(0,σ) is a normally distributed random variable with a mean of 0 and a variance of σ. 2 ;P n It is the noise floor.

5. The event-driven dynamic sorting activation and information fusion method for sensor networks according to claim 1, characterized in that, In step 4, the state ranking values ​​of sensor nodes are calculated based on the important parameters of sensor nodes in the topology network energy model and the link quality model. Joint parameters for the event-driven region and activation ratio in the ranking activation method are designed, and the sensor nodes are activated in conjunction with the state ranking values, as detailed below: Step 4.1: Establish a dynamic status sorting mechanism for sensor nodes within the region: The network simulates the occurrence of events based on the Poisson model. Due to the self-driven wake-up switch, sensor nodes within a certain distance will be temporarily physically woken up after an event occurs. At the beginning, each sensor node sends its important parameters to the fusion center, including the remaining energy of the sensor node, the distance between sensor nodes and the link quality, and the fusion center maintains and updates them. Within the event area, the initial state sorting value R for each sensor node is calculated using the following formula. i : Where, γ ij Distance represents the link quality between sensor node i and its neighbor node j. ij SNE represents the Euclidean distance between sensor node i and its neighbor node j; i is the remaining energy of the sensor node; N is the number of sensor nodes in the entire directed graph network topology; The formula for calculating the sorted value of sensor node i in the (n+1)th iteration as follows: Among them, DEG ij α is the matrix of outer links of sensor node i's neighbor nodes j; α is the damping factor; Step 4.2: Design the joint parameters of the event-driven region and activation ratio in the activation method, including the area S of the activation region and the activation ratio J of the sensor nodes within that region, and activate the sensor nodes in combination with the state sorting value: In the sensor node state sorting mechanism, all surviving nodes within the circular region S surrounding the event can be triggered, thus avoiding unnecessary activation of sensor nodes that are too far from the event. Before activating nodes, the selection ratio of sensor nodes is determined. J = num / ρS (9) Where J is the activation ratio of sensor nodes in an area of ​​S; ρ is the spatial distribution density of sensor nodes in the topology network; and num is the minimum number of activated nodes required to achieve the specified monitoring accuracy. In each working cycle of the sensor node, by combining the sensor status sorting value and joint parameters, a suitable sensor node is selected to remain active and to accurately monitor the target or event.

6. The event-driven dynamic sorting activation and information fusion method for sensor networks according to claim 1, characterized in that, In step 5, the awakened sensor nodes perform information fusion using an improved distributed Kalman fusion method, introducing weights driven by link quality. Each sensor node updates its own estimated data based on its own estimated data, neighbor's estimated data, and the weight matrix to obtain an accurate target state estimate, as detailed below: Step 5.1: Based on the logarithmic signal path attenuation model in Step 3, introduce weights driven by link quality and update the weights in real time: Extend the graph G = {ν, ε} to Among them, the weight matrix n is the dimension of the target state, and the link quality calculated according to the logarithmic signal path attenuation model represents the ability of sensor nodes to transmit data to each other. Step 5.2: Each sensor node updates its own estimated data based on its own estimated data, neighbor estimated data, and weight matrix: For sensor nodes within the activated area, the linear discrete-time dynamic model of the wireless sensor topology network is: x k+1 =Ax k +Bw k (12) Where, x k This represents the system state information at time k; Let A represent the measurement value of sensor node i at time k; let B represent the state transition matrix; let H represent the noise matrix of the system input; and let H represent the measurement value of sensor node i at time k. i The measurement matrix representing sensor node i; w k This represents the process noise at time k. This represents the measurement noise of sensor node i at time k; Set an initial measurement information transfer matrix and a state information transfer matrix in, It is the measurement matrix of sensor node i at time k; It is the measurement error covariance matrix; Each sensor node aggregates information from all active nodes in the entire topology network; considering the impact of link quality on the reliability of information transmission, link quality-driven weights are introduced during the information fusion process, and these weights are updated in real time; for sensor node i, it undergoes iteration d a Next, d a Given the maximum number of hops in the network topology, the following summary results are obtained; For sensor node i, the t-th iteration is performed as follows: in, The weight elements in the ω satisfy ω ij =ω ji >0 and ω ii =1; t=1,2,3,...,d a ; Step 5.3: Each sensor node obtains an accurate target state estimate through prediction and updating, and achieves finite-time convergence of the fusion error: After d a The next iteration yields the extended graph. From layer 0 to d a The sensor node sensing information transmission matrix α of the layer i (d a ) and Q i (d a In the information fusion process, the following formulas were used for prediction and updating: predict: renew: in, The prior state estimate of sensor node i at time k; Let be the prior estimation error covariance of sensor node i at time k; The posterior state estimate of sensor node i at time k; Let F be the covariance of the posterior estimation error at time k; k-1 Given the system matrix; Y k-1 Here is the process error covariance matrix; T denotes transpose; The final fusion result with the minimum variance is obtained by fusing the information referenced by each sensor node using the optimal information fusion calculation method. The formula for the information fusion calculation method is as follows: Where, Γ=Σ -1 e(e T Σ -1 e) -1 It is a positive definite matrix. Let n be the cross-covariance matrix, n be the target state dimension, and l be the number of sensors. It is the local process noise covariance matrix of sensor node i and its neighbor node j.

Images