An intelligent clustering method for power transmission corridor sensor network

Abstract

The present invention discloses an intelligent clustering method for a power transmission corridor sensor network. The candidate cluster head node calculates the evaluation value of the candidate cluster head according to the sensor deployment location, residual energy, distribution density and weight of influencing parameters, and calculates its own service radius according to the evaluation value of the candidate cluster head. The candidate cluster head node competes with the neighbor candidate cluster head node for the formal cluster head node according to the service radius to obtain the formal cluster head node set; the formal cluster head node in the formal cluster head node set selects the member node to be scheduled according to its own residual energy and wireless resource status to form a non-uniform cluster; the formal cluster head node allocates time slots to the member nodes in the non-uniform cluster for intra-cluster data transmission. The present invention targets the underground power transmission corridor scenario, and based on the method combining intelligent twins with actual networks, the sensor equipment is clustered and managed, thereby reducing the sensor energy consumption while ensuring the performance requirements of the perceived data transmission.

Description

Technical Field

[0001] The invention relates to an intelligent clustering method for a power transmission corridor sensor network, belonging to the technical field of wireless sensor networks. Background Art

[0002] Urban underground comprehensive pipeline corridors play an irreplaceable role in the process of modern urban construction, which can not only alleviate traffic congestion, but also provide convenience for communication, drainage and power transmission. In particular, the power transmission pipeline corridor is equipped with inspection and maintenance space, and laying, addition, reduction and maintenance can be carried out directly in the pipeline corridor, which greatly reduces the cost of multiple road renovations and the maintenance cost of engineering pipelines.

[0003] At present, the inspection and troubleshooting of pipeline corridors mostly rely on manual work, and there is a lack of mature, efficient, low-cost, and full-area coverage pipeline corridor operation environment and status monitoring and automatic early warning solutions. With the continuous deepening of the construction of the power Internet of Things and the increasing number of transmission pipeline corridors, the demand for transmission pipeline corridor monitoring will become increasingly urgent, and low-cost, high-efficiency digital operation and maintenance means are urgently needed. Wireless sensor networks have the lowest cost, and with the development of low-power wireless sensor network technology, their power supply requirements are getting lower and lower, making them one of the most economical and applicable ways to cover pipeline corridors.

[0004] Due to the limited energy of wireless sensors, in order to extend the life of wireless sensor networks, sensor clustering is usually used to transmit sensor data. However, due to the complex environment and narrow structure inside the transmission corridor, the existing ground wireless sensor network clustering method will face the dilemma of low efficiency.

[0005] The existing ground wireless sensor clustering method is based on uniform clustering, which is difficult to adapt to the narrow and long terrain characteristics of the power transmission corridor. Uniform clustering will inevitably cause excessive energy consumption or excessive data load of the sensor cluster heads around the base station, resulting in a shortened sensor life and affecting the long-term reliable operation of the entire sensor network. In addition, the existing wireless sensor clustering process needs to comprehensively consider many factors such as residual energy consumption and communication distance. However, there is a lack of limited guidance mechanism for the weight setting of such factors. Summary of the invention

[0006] Objective: In order to overcome the problems existing in the prior art of how to determine the factors affecting the performance of sensor networks and their influence weights, how to reasonably perform non-uniform clustering of sensors, and how to reasonably realize inter-cluster transmission of perceived data, the present invention provides an intelligent clustering method for a transmission corridor sensor network. Aiming at the underground transmission corridor scenario, the sensor equipment is clustered and managed based on a method combining smart twins with actual networks, thereby reducing sensor energy consumption while ensuring the performance requirements of perceived data transmission.

[0007] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for intelligent clustering of a power transmission corridor sensor network comprises the following steps:

[0009] In a pre-set intelligent twin network, a twin model for calculating the weights of influencing parameters is obtained based on a proximal strategy optimization method, and the weights of influencing parameters are obtained according to the twin model for calculating the weights of influencing parameters.

[0010] At the beginning of each clustering cycle, each sensor generates a random number. When the random number is greater than a fixed threshold, it is selected as a candidate cluster head node. Otherwise, the node is an ordinary node during the clustering cycle.

[0011] The candidate cluster head node calculates the evaluation value of the candidate cluster head according to the sensor deployment location, residual energy, distribution density and influencing parameter weights, and calculates its own service radius according to the evaluation value of the candidate cluster head. The candidate cluster head node competes with its neighbor candidate cluster head nodes for the formal cluster head node based on the service radius to obtain the formal cluster head node set.

[0012] The formal cluster head node in the formal cluster head node set selects the member nodes for scheduling according to its own remaining energy and wireless resource status to form a non-uniform cluster.

[0013] The formal cluster head node allocates time slots to the member nodes in the non-uniform cluster for intra-cluster data transmission.

[0014] As a preferred solution, the smart twin network is constructed using Monte Carlo simulation based on the deployment location, residual energy and distribution density of sensors in the actual transmission corridor, and is used to simulate the actual transmission corridor sensor network.

[0015] As a preferred solution, the method for obtaining the twin model for calculating the weight of influencing parameters based on the method of proximal strategy optimization includes the following steps:

[0016] Step 1: Initialize the remaining energy, deployment location, and distribution density of the sensor network, and obtain the state transition probability Pr{E n (t+1),L n (t+1),C n (t+1)}|{E n (t+1),L n (t+1),C n (t+1)},w E (t),w L (t),w C (t)}, stored in the state cache unit.

[0017] Step 2: Based on the state transition probability, the state, action, reward and next cycle state prediction parameters {s, a, r, s_} in the Markov decision process are stored in the model memory.

[0018] Step 3: Rewards are stored in chronological order from the model memory according to the time decay rate γ -t After accumulation, it is input into the evaluation network to obtain the expected cumulative return v and the expected state-action function q.

[0019] Step 4: Construct qv as the advantage function and train the evaluation network based on the mean square error.

[0020] Step 5: Copy the evaluation network to become a new action network, input the current state into the new action network and the old action network respectively, obtain the new and old normal distributions of the sensor network state for the next clustering period, and use the new and old normal distributions divided by each other as the advantage function to further train the new action network until the new action network converges.

[0021] As a preferred solution, the Markov decision process acquisition method is as follows:

[0022] Construct a Markov decision process with state space S, behavior space A and reward function r:S×S×A→R, where the parameters are defined as follows:

[0023] State space S: includes the residual energy parameters {E n (t), n=1,2,...,N(t)}, placement position parameter {L n (t), n=1,2,...,N(t)} and the distribution density parameter {C n (t),n=1,2,...,N(t)}.

[0024] Action space A: including the residual energy parameters of the sensor nodes affecting the weight w E (t), the location parameter influences the weight w L (t) and the distribution density parameter affects the weight w C (t).

[0025] Reward function r: S×S×A→R, where R represents the set of all values ​​of the reward function r. When the user is in state s, s∈S, and the smart twin network takes action a, a∈A, the reward function can be calculated by the following formula:

[0026]

[0027] in, Represents the total amount of remaining energy in the network.

[0028] As a preferred solution, the node that served as the formal cluster head node in the previous clustering cycle serves as an ordinary node in this round.

[0029] As a preferred solution, the deployment position, residual energy and distribution density of the sensor are calculated as follows:

[0030]

[0031] Where: E n (t) represents sensor s n The remaining energy level in the current sensor network, T represents the number of sensor clustering cycles expected by the sensor network, e n (t) represents sensor s n The residual energy of , t represents the t-th clustering period, Represents the average sensor energy of the entire transmission corridor during the initial clustering period.

[0032]

[0033] Among them, L n (t) represents sensor s n The deployment location parameter in the current sensor network, l N(t) For sensors N(t) Distance to the base station, l n For sensors n Distance to the base station.

[0034]

[0035] Among them, C n (t) is the sensor distribution density parameter, c n (t) Sensors n The number of neighbor nodes within the communication range.

[0036] As a preferred solution, the evaluation value calculation formula of the candidate cluster head is as follows:

[0037] I m (t) = w E (t)E n (t)+w L (t)L n (t)+w C (t)C n (t)

[0038] Among them, I m (t) represents the candidate cluster head v m The evaluation value, w E (t) represents the residual energy parameter influence weight, w L (t) represents the influence weight of the deployment location parameter, wC (t) represents the influence weight of distribution density parameter.

[0039] As a preferred solution, it also includes constraints that affect the weight parameters. The constraint calculation formula is as follows:

[0040] w E (t)+w L (t)+w C (t)=1.

[0041] As a preferred solution, the service radius calculation formula is as follows:

[0042] D m (t) = θI m (t)R

[0043] Where: R represents the maximum competition radius pre-designed by the candidate cluster head, θ is the empirical correction factor, I m (t) represents the candidate cluster head v m The evaluation value.

[0044] As a preferred solution, the method of calculating the service radius of the candidate cluster head according to the evaluation value of the candidate cluster head, the candidate cluster head node competes with the neighbor candidate cluster head node for the formal cluster head node according to the service radius, and obtaining the formal cluster head node set includes the following steps:

[0045] According to each candidate cluster head v m Service radius D m (t) Determine its potential service scope m (t), if there exists G m (t)∈G m' (t), m≠m', then the candidate cluster head v m In this round of clustering cycle, it serves as a common node. Otherwise, it serves as a formal cluster head node. Finally, the formal cluster head node set {b1, b2, ···, b m ,···,b K(t)}, where K(t) is the number of clusters.

[0046] As a preferred solution, the method in which the formal cluster head node in the formal cluster head node set selects to schedule the member nodes according to its own residual energy and wireless resource status to form a non-uniform cluster includes the following steps:

[0047] Step 1: Formal cluster head node b m The transmission power of broadcast information is adjusted according to its service radius, and the cluster head identity is broadcast to all sensors within its service radius. Member nodes select the nearest formal cluster head node to join.

[0048] Step 2: The formal cluster head node selects member nodes according to its own maximum data carrying capacity and the order in which the member request to join the cluster is received. If all member nodes successfully join the cluster, execute step 4. If the number of members exceeds the maximum carrying capacity of the formal cluster head node, and sends a rejection message to the unselected members, execute step 3.

[0049] Step 3: The unselected members send a request message to the nearest formal cluster head node except the rejected formal cluster head node, requesting to join the sub-cluster, and return to step 2.

[0050] Step 4: Clustering is completed.

[0051] As a preferred solution, the formal cluster head node allocates data transmission resources to member nodes in a time division multiplexing mode.

[0052] Beneficial effects: The intelligent clustering method of a power transmission corridor sensor network provided by the present invention has the following advantages:

[0053] 1. According to the residual energy, deployment location and distribution density of the nodes, the problem of non-uniform clustering of the transmission corridor sensor network is solved. The influencing parameter weight analysis method based on proximal strategy optimization and the non-uniform cluster head election method are conducive to energy saving of the transmission corridor sensor network and achieve the purpose of extending the network life.

[0054] 2. Based on the smart twin model of the proximal optimization strategy and the calculation method of the influencing parameter weights, the residual energy parameters, deployment location parameters and distribution density parameters of the sensor are obtained, which is helpful to select the sensor cluster head node more scientifically. According to the reward function of the smart twin model, the reward and weight of the residual energy and the penalty of deviation are covered, which can more effectively optimize the weights of the influencing parameters. In addition, the proposed proximal optimization strategy network model has low sensitivity to neural network hyperparameters, fast training convergence and stable training results, which is more conducive to actual engineering deployment.

[0055] 3. Based on the cluster head cluster radius and service range, the cluster head selection can be quickly determined while ensuring the coverage of ordinary nodes; the bidirectional selection mechanism of member nodes and cluster head nodes can ensure the energy balance between cluster heads. In addition, the non-uniform cluster heads formed by the proposed method appropriately reduce the clustering scale when the equipment is close to the base station, the residual energy is large, and the distribution density is large, which is conducive to reducing the energy consumption of the cluster head and improving the overall network life of the transmission corridor sensor network. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0057] The present invention will be further described below in conjunction with specific embodiments.

[0058] like Figure 1 As shown, a method for intelligent clustering of a power transmission corridor sensor network includes the following steps:

[0059] According to the deployment location, residual energy and distribution density of sensors in the actual transmission corridor, a smart twin network is constructed using Monte Carlo simulation to simulate the actual transmission corridor sensor network. In the smart twin network, a twin model for calculating the influencing parameter weights is obtained based on the proximal strategy optimization method, and the influencing parameter weights are obtained based on the twin model.

[0060] At the beginning of each clustering cycle, each sensor generates a random number. When the random number is greater than a fixed threshold, it is selected as a candidate cluster head node. Otherwise, the node is an ordinary node during the clustering cycle.

[0061] The candidate cluster head node calculates the evaluation value of the candidate cluster head according to the sensor deployment location, residual energy, distribution density and influencing parameter weights, and calculates its own service radius according to the evaluation value of the candidate cluster head. The candidate cluster head node competes with its neighbor candidate cluster head nodes for the formal cluster head node based on the service radius to obtain the formal cluster head node set.

[0062] The formal cluster head node in the formal cluster head node set selects the member nodes for scheduling according to its own remaining energy and wireless resource status to form a non-uniform cluster.

[0063] The formal cluster head node allocates time slots to the member nodes in the non-uniform cluster for intra-cluster data transmission.

[0064] The method for obtaining the deployment position, residual energy and distribution density of sensors in the actual power transmission corridor comprises the following steps:

[0065] Defined as follows: At the beginning of the t-th clustering period, the sensor set in the transmission corridor is represented as {s1,s2,···,s n ,···,s N(t)}, where s n represents the nth sensor in the current clustering period, and N(t) represents the number of sensors in the current clustering period.

[0066] Since the energy storage of sensors is limited and the replacement cost is high, clustering is based on extending the working life of sensors as much as possible, that is, extending the number of sensor clustering cycles as much as possible. The present invention comprehensively considers the impact of sensor deployment location, residual energy and distribution density on the clustering performance of the sensor network, designs a smart twin network, and quantifies the influence weights of the three parameters.

[0067] (1) Sensor remaining energy parameters

[0068] In order to make all sensors run out of energy at the same time and maximize the overall life of the sensor network, sensors with more remaining energy should be selected as formal cluster head nodes. n The remaining energy is e n (0), then the average sensor energy of the entire transmission corridor initial clustering period is It can be calculated by the following formula:

[0069]

[0070] Wherein, N(0) represents the number of sensor nodes in the initial clustering period.

[0071] At the beginning of the t-th clustering period, e n (t) represents sensor s n Due to the large number of sensor network nodes, it is very expensive to obtain the residual energy of all sensor nodes in each clustering cycle. n The residual energy parameter is calculated by the following formula:

[0072]

[0073] Where: E n (t) represents sensor s n In the current residual energy level of the sensor network, T represents the number of sensor clustering cycles expected by the sensor network. The larger the residual energy parameter value of the sensor, the higher the probability that the sensor will be selected as the formal cluster head node.

[0074] (2) Sensor deployment location parameters

[0075] Since the cluster head node closer to the base station can complete data transmission with lower energy consumption, the sensor closer to the base station should be selected as the formal cluster head node as much as possible.

[0076] Definition n Indicates sensor n The distance to the base station. The sensor location parameter represents the distance from the sensor to the base station. It is calculated as follows:

[0077]

[0078] Among them, L n (t) represents sensor s n The deployment location parameter in the current sensor network, l N(t) For sensors N(t) Distance to the base station, l n For sensors nThe distance to the base station, the larger the value of the sensor deployment location parameter is, the closer it is to the base station, and the higher the probability of being selected as the formal cluster head node.

[0079] (3) Sensor distribution density parameters

[0080] Considering the narrow and long structure characteristics of the transmission corridor and the energy consumption requirements of sensors, more nodes should be selected as cluster heads in the node-dense area to reduce the size of the cluster and reduce the energy consumption of the formal cluster head nodes; in the node-sparse area, the probability of a node becoming a formal cluster head node should be appropriately reduced to balance the energy consumption of the formal cluster head nodes of the transmission corridor sensor network. At the beginning of the t-th clustering cycle, define c n (t) is sensor s n The number of neighbor nodes within the communication range and the sensor distribution density parameter C n (t) can be calculated by the following formula:

[0081]

[0082] Among them, the larger the value of the sensor distribution density parameter is, the greater the sensor distribution density in the area is, and the higher the probability that the sensor in the area is selected as the formal cluster head is.

[0083] Since the energy and computing power of actual physical network sensors are limited, the present invention simulates the actual transmission corridor sensor network by Monte Carlo simulation in the smart twin network, and obtains the influence parameter weight calculation twin model based on the proximal strategy optimization method. The influence parameter weights are obtained through the influence parameter weight calculation twin model, thereby guiding the actual transmission corridor sensor network to update the influence parameter weights.

[0084] The method for obtaining the twin model for calculating the weight of influencing parameters based on the proximal strategy optimization method includes the following steps:

[0085] Step 1: Initialize the remaining energy, deployment location, and distribution density of the sensor network, and obtain the state transition probability Pr{E n (t+1),L n (t+1),C n (t+1)}|{E n (t+1),L n (t+1),C n (t+1)},w E (t),w L (t),w C (t)}, stored in the state cache unit.

[0086] The Markov decision process acquisition method is as follows:

[0087] Construct a Markov decision process with state space S, behavior space A and reward function r:S×S×A→R, where the parameters are defined as follows:

[0088] State space S: includes the residual energy parameters {E n (t), n=1,2,...,N(t)}, placement position parameter {L n (t), n=1,2,...,N(t)} and the distribution density parameter {C n (t),n=1,2,...,N(t)}.

[0089] Action space A: including the residual energy parameters of the sensor nodes affecting the weight w E (t), the location parameter influences the weight w L (t) and the distribution density parameter affects the weight w C (t).

[0090] Reward function r: S×S×A→R, where R represents the set of all values ​​of the reward function r. When the user is in state s, s∈S, and the smart twin network takes action a, a∈A, the reward function can be calculated by the following formula:

[0091]

[0092] in, Represents the total amount of remaining energy in the network. The larger the total amount, the greater the reward obtained by the twin model that affects the parameter weight calculation. E (t)+w L (t)+w C (t)-1) 2 It means that the closer the sum of weight constraints is to 1, the greater the reward obtained by the twin model that affects the parameter weight calculation.

[0093] Step 2: Based on the state transition probability obtained in step 1, the state, action, reward and next cycle state prediction parameters {s, a, r, s_} in the Markov decision process are stored in the model memory.

[0094] Step 3: Rewards are stored in chronological order from the model memory according to the time decay rate γ -t After accumulation, it is input into the evaluation network to obtain the expected cumulative return v and the expected state-action function q.

[0095] Step 4: Construct qv as the advantage function and train the evaluation network based on the mean square error.

[0096] Step 5: Copy the evaluation network to become a new action network, input the current state into the new action network and the old action network respectively, obtain the new and old normal distributions of the sensor network state for the next clustering period, and use the new and old normal distributions divided by each other as the advantage function to further train the new action network until the new action network converges.

[0097] In order to prevent some nodes from acting as cluster heads too often and speed up cluster head election, at the beginning of each clustering cycle:

[0098] (1) To avoid excessive energy consumption of a single cluster head, the node that served as the formal cluster head node in the previous round of clustering serves as an ordinary node in this round;

[0099] (2) Each sensor that served as an ordinary node in the previous round generates a random number p∈[0,1] locally. When p is greater than the threshold p0 set by the sensor network, it participates in the cluster head competition. Otherwise, it serves as an ordinary node in the clustering cycle. Finally, a list of candidate cluster head nodes {v1,v2,···,v m ,···,v M(t)}, where M(t) represents the number of candidate cluster heads in the tth clustering period.

[0100] For each candidate cluster head, it needs to compete with other candidate cluster heads for its service radius. Since the candidate cluster heads with more residual energy and closer to the base station must bear more data forwarding, in order to balance the energy consumption of the cluster heads, such cluster heads should be configured with a smaller service radius. In summary, the cluster head service radius can be calculated by the following formula:

[0101] D m (t) = θI m (t)R (5)

[0102] Where: R represents the maximum competition radius pre-designed by the candidate cluster head, θ is the empirical correction factor, I m (t) represents the candidate cluster head v m The evaluation value of I m (t) The larger the value, the m The greater the probability of becoming a cluster head.

[0103] The calculation method of the evaluation value of the sensor candidate cluster head is as follows:

[0104] After obtaining the deployment location, residual energy, and distribution density of sensors in the actual transmission corridor, the evaluation value of the sensor candidate cluster head including the influence weight is constructed as a reference for the candidate cluster head node to elect the formal cluster head node. The specific calculation method is as follows:

[0105] I m (t) = w E (t)E n(t)+w L (t)L n (t)+w C (t)C n (t) (6)

[0106] Among them, I m (t) represents the candidate cluster head v m The evaluation value of I m (t) The larger the value, the m The greater the probability of becoming a formal cluster head node. E (t) represents the residual energy parameter influence weight, w L (t) represents the influence weight of the deployment location parameter, w C (t) represents the distribution density parameter affecting the weight, and the constraints of the three weight parameters are calculated by the following formula:

[0107] w E (t)+w L (t)+w C (t)=1 (7)

[0108] In order to reduce the complexity of weight optimization and ensure the real-time performance of clustering optimization of the transmission corridor sensor network, w E (t), w L (t), w C The value space of (t) is limited to the set {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1}.

[0109] According to each candidate cluster head v m Service radius D m (t) Determine its potential service scope m (t), if there exists G m (t)∈G m' (t), m≠m', then the candidate cluster head v m In this round of clustering cycle, it serves as a common node. Otherwise, it serves as a formal cluster head node. Finally, the formal cluster head node set {b1, b2, ···, b m ,···,b K(t)}, where K(t) is the number of clusters.

[0110] After the formal cluster head node is determined, the clustering process is as follows:

[0111] Step 1: Formal cluster head node b m The transmission power of broadcast information is adjusted according to its service radius, and the cluster head identity is broadcast to all sensors within its service radius. Member nodes select the nearest formal cluster head node to join.

[0112] Step 2: The formal cluster head node selects member nodes according to its own maximum data carrying capacity and the order in which the member request to join the cluster is received. If all member nodes successfully join the cluster, execute step 4. If the number of members exceeds the maximum carrying capacity of the formal cluster head node, and sends a rejection message to the unselected members, execute step 3.

[0113] Step 3: The unselected members send a request message to the nearest formal cluster head node except the rejected formal cluster head node, requesting to join the sub-cluster, and return to step 2.

[0114] Step 4: After clustering is completed, the cluster head node allocates data transmission resources to member nodes in a time division multiplexing mode.

[0115] Example:

[0116] The method of the present invention can also bring the following benefits during use:

[0117] The algorithm operation in the clustering process of the physical sensor network can be transferred to the twin model simulation calculation, which not only effectively reduces the computing and communication overhead of the physical sensor network and extends the network life, but also can provide predictions of relevant parameter weights for the sensor network clustering and accelerate the completion of clustering.

[0118] It can solve the problem of non-uniform clustering of the transmission corridor sensor network. The non-uniform cluster head election method is beneficial to the energy saving of the transmission corridor sensor network and achieves the purpose of extending the network life.

[0119] The clustering parameter modeling of the transmission corridor sensor network comprehensively considers the influence of sensor residual energy, location information and distribution density on clustering. Due to the large number of sensors, real-time residual energy statistics of all sensors will seriously reduce the running rate of the clustering algorithm. Based on this, the present invention proposes an estimation method for the residual energy of all sensors in the network, namely, Where t represents the period of the current cluster, T represents the maximum number of periods that the sensor network is expected to use, Represents the initial total energy of the network

[0120] The proximal policy optimization method constructs a reward function This reward function covers the reward and weight of the remaining energy and the penalty for deviation from 1, which can more effectively optimize the weight of the influencing parameters. In addition, the proposed proximal optimization strategy network model has low sensitivity to neural network hyperparameters, fast training convergence, stable training results, etc., which is more conducive to actual engineering deployment.

[0121] By determining the cluster head cluster radius and service range, the cluster head selection can be quickly determined while ensuring the coverage of ordinary nodes; the bidirectional selection mechanism of member nodes and cluster head nodes can ensure the energy balance between cluster heads. In addition, the non-uniform cluster heads formed by the proposed method appropriately reduce the clustering scale when the equipment is close to the base station, the residual energy is large, and the distribution density is large, which is conducive to reducing the energy consumption of the cluster head and improving the overall network life of the transmission corridor sensor network.

[0122] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for intelligent clustering of a power transmission corridor sensor network, characterized in that: The steps include: In a pre-set smart twin network, a twin model for calculating the weight of an influencing parameter is obtained based on a proximal strategy optimization method, and the weight of the influencing parameter is obtained according to the twin model for calculating the weight of the influencing parameter; At the beginning of each clustering cycle, each sensor generates a random number. When the random number is greater than a fixed threshold, it is selected as a candidate cluster head node. Otherwise, the node is an ordinary node in the clustering cycle. The candidate cluster head node calculates the evaluation value of the candidate cluster head according to the sensor deployment location, residual energy, distribution density and influencing parameter weights, and calculates its own service radius according to the evaluation value of the candidate cluster head. The candidate cluster head node competes with the neighbor candidate cluster head nodes for the formal cluster head node according to the service radius to obtain the formal cluster head node set; The formal cluster head node in the formal cluster head node set selects the member nodes for scheduling according to its own remaining energy and wireless resource status to form a non-uniform cluster; The formal cluster head node allocates time slots to the member nodes in the non-uniform cluster for intra-cluster data transmission; The method in which the formal cluster head node in the formal cluster head node set selects to schedule member nodes according to its own remaining energy and wireless resource status to form a non-uniform cluster comprises the following steps: Step 1: Formal cluster head node b m Adjust the transmission power of broadcast information according to its service radius, and broadcast the cluster head identity to all sensors within its service radius. Member nodes select the nearest formal cluster head node to join; Step 2: The formal cluster head node selects member nodes according to its own maximum data carrying capacity and the order in which the member request to join the cluster is received; if all member nodes successfully join the sub-cluster, execute step 4; if the number of members exceeds the maximum carrying capacity of the formal cluster head node, and sends a rejection message to the unselected members, execute step 3; Step 3: The unselected members send a request message to the nearest formal cluster head node except the rejected formal cluster head node, requesting to join the sub-cluster, and return to step 2; Step 4: Clustering is completed.

2. According to claim 1, a method for intelligent clustering of a power transmission corridor sensor network is characterized in that: The smart twin network is constructed using Monte Carlo simulation based on the deployment location, residual energy and distribution density of sensors in the actual transmission corridor, and is used to simulate the actual transmission corridor sensor network.

3. According to claim 1, a method for intelligent clustering of a power transmission corridor sensor network is characterized in that: The method for obtaining the twin model for calculating the weight of influencing parameters based on the method of proximal strategy optimization comprises the following steps: Step 1: Initialize the remaining energy, deployment location, and distribution density of the sensor network, and obtain the state transition probability Pr{E n (t+1),L n (t+1),C n (t+1)}|{E n (t+1),L n (t+1),C n (t+1)},w E (t),w L (t),w C (t)}, stored in the state cache unit; Step 2: Based on the state transition probability, the state, action, reward and next cycle state prediction parameters {s, a, r, s_} in the Markov decision process are stored in the model memory; Step 3: Rewards are stored in chronological order from the model memory according to the time decay rate γ -t After accumulation, it is input into the evaluation network to obtain the expected cumulative return v and the expected state-action function q; Step 4: Construct qv as the advantage function and train the evaluation network based on the mean square error; Step 5: Copy the evaluation network to become a new action network, input the current state into the new action network and the old action network respectively, obtain the new and old normal distributions of the sensor network state for the next clustering period, and use the new and old normal distributions divided by each other as the advantage function to further train the new action network until the new action network converges.

4. The method for intelligent clustering of a power transmission corridor sensor network according to claim 3 is characterized by: The Markov decision process acquisition method is as follows: Construct a Markov decision process with state space S, behavior space A and reward function r:S×S×A→R, where the parameters are defined as follows: State space S: includes the residual energy parameters {E n (t), n=1,2,...,N(t)}, placement position parameter {L n (t), n=1,2,...,N(t)} and the distribution density parameter {C n (t), n=1,2,...,N(t)}; Action space A: including the residual energy parameters of the sensor nodes affecting the weight w E (t), the location parameter influences the weight w L (t) and the distribution density parameter affects the weight w C (t); Reward function r: S×S×A→R, where R represents the set of all values ​​of the reward function r. When the user is in state s, s∈S, and the smart twin network takes action a, a∈A, the reward function can be calculated by the following formula: in, Represents the total amount of remaining energy in the network.

5. The method for intelligent clustering of a power transmission corridor sensor network according to claim 1, characterized in that: The node that served as the formal cluster head node in the previous clustering cycle serves as an ordinary node in this round.

6. The method for intelligent clustering of a power transmission corridor sensor network according to claim 1, characterized in that: The deployment position, remaining energy and distribution density of the sensor are calculated as follows: Where: E n (t) represents sensor s n The remaining energy level in the current sensor network, T represents the number of sensor clustering cycles expected by the sensor network, e n (t) represents sensor s n The residual energy of , t represents the t-th clustering period, represents the average energy of sensors during the initial clustering period of the entire transmission corridor; Among them, L n (t) represents sensor s n The deployment location parameter in the current sensor network, l N(t) For sensors N(t) Distance to the base station, l n For sensors n Distance to the base station; Among them, C n (t) is the sensor distribution density parameter, c n (t) Sensors n The number of neighbor nodes within the communication range.

7. The method for intelligent clustering of a power transmission corridor sensor network according to claim 6, characterized in that: The evaluation value calculation formula of the candidate cluster head is as follows: I m (t)=w E (t)E n (t)+w L (t)L n (t)+w C (t)C n (t) Among them, I m (t) represents the candidate cluster head v m The evaluation value, w E (t) represents the residual energy parameter influence weight, w L (t) represents the influence weight of the deployment location parameter, w C (t) represents the influence weight of distribution density parameter.

8. The method for intelligent clustering of a power transmission corridor sensor network according to claim 7, characterized in that: It also includes constraints that affect the weight parameters. The constraint calculation formula is as follows: w E (t)+w L (t)+w C (t)=1。 9. The method for intelligent clustering of a power transmission corridor sensor network according to claim 1, characterized in that: The service radius calculation formula is as follows: D m (t)=θI m (t)R Where: R represents the maximum competition radius pre-designed by the candidate cluster head, θ is the empirical correction factor, I m (t) represents the candidate cluster head v m The evaluation value.

10. The method for intelligent clustering of a power transmission corridor sensor network according to claim 1, characterized in that: The method of calculating the service radius of the candidate cluster head according to the evaluation value of the candidate cluster head, and the candidate cluster head node competes with the neighbor candidate cluster head node for the formal cluster head node according to the service radius to obtain the formal cluster head node set includes the following steps: According to each candidate cluster head v m Service radius D m (t) Determine its potential service scope m (t), if there exists G m (t)∈G m' (t), m≠m', then the candidate cluster head v m In this round of clustering cycle, it serves as a common node. Otherwise, it serves as a formal cluster head node. Finally, the formal cluster head node set {b1, b2, ···, b m ,···,b K(t) }, where K(t) is the number of clusters.

11. The method for intelligent clustering of a power transmission corridor sensor network according to claim 1, characterized in that: The formal cluster head node allocates data transmission resources to member nodes in a time division multiplexing mode.

Images